Upload folder using huggingface_hub

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ2_M.layout.json

@@ -0,0 +1,89 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ2_M",
+  "base_type": "iq2_s",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0911,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0569,
+      "type": "q3_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q3_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 5860008680,
+    "map_body+embd": 5545681640,
+    "ratio": 0.9464,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 2.833,
+  "body_bpw_heuristic": 3.023,
+  "total_params": 14765593600,
+  "file_bytes_pred": 5547381480,
+  "file_bpw_pred": 3.006,
+  "histogram": {
+    "attn_k": {
+      "iq4_xs": 39,
+      "q4_k": 1
+    },
+    "attn_output": {
+      "iq4_xs": 29,
+      "iq2_s": 10,
+      "iq3_xxs": 1
+    },
+    "attn_q": {
+      "iq2_s": 40
+    },
+    "attn_v": {
+      "q5_k": 30,
+      "iq4_xs": 8,
+      "q4_k": 2
+    },
+    "ffn_down": {
+      "iq2_s": 30,
+      "iq3_xxs": 8,
+      "iq3_s": 2
+    },
+    "ffn_gate": {
+      "iq2_s": 40
+    },
+    "ffn_up": {
+      "iq2_s": 30,
+      "iq3_s": 9,
+      "iq4_xs": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ2_M",
+  "quantize_ftype": "IQ2_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ2_M.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q3_k
+^blk\.0\.attn_k\.weight$=q4_k
+^blk\.0\.attn_output\.weight$=iq3_xxs
+^blk\.0\.attn_q\.weight$=iq2_s
+^blk\.0\.attn_v\.weight$=q4_k
+^blk\.0\.ffn_down\.weight$=iq2_s
+^blk\.0\.ffn_gate\.weight$=iq2_s
+^blk\.0\.ffn_up\.weight$=iq2_s
+^blk\.1\.attn_k\.weight$=iq4_xs
+^blk\.1\.attn_output\.weight$=iq4_xs
+^blk\.1\.attn_q\.weight$=iq2_s
+^blk\.1\.attn_v\.weight$=q5_k
+^blk\.1\.ffn_down\.weight$=iq3_s
+^blk\.1\.ffn_gate\.weight$=iq2_s
+^blk\.1\.ffn_up\.weight$=iq3_s
+^blk\.2\.attn_k\.weight$=iq4_xs
+^blk\.2\.attn_output\.weight$=iq4_xs
+^blk\.2\.attn_q\.weight$=iq2_s
+^blk\.2\.attn_v\.weight$=q5_k
+^blk\.2\.ffn_down\.weight$=iq3_xxs
+^blk\.2\.ffn_gate\.weight$=iq2_s
+^blk\.2\.ffn_up\.weight$=iq3_s
+^blk\.3\.attn_k\.weight$=iq4_xs
+^blk\.3\.attn_output\.weight$=iq4_xs
+^blk\.3\.attn_q\.weight$=iq2_s
+^blk\.3\.attn_v\.weight$=q5_k
+^blk\.3\.ffn_down\.weight$=iq3_xxs
+^blk\.3\.ffn_gate\.weight$=iq2_s
+^blk\.3\.ffn_up\.weight$=iq3_s
+^blk\.4\.attn_k\.weight$=iq4_xs
+^blk\.4\.attn_output\.weight$=iq4_xs
+^blk\.4\.attn_q\.weight$=iq2_s
+^blk\.4\.attn_v\.weight$=q5_k
+^blk\.4\.ffn_down\.weight$=iq3_xxs
+^blk\.4\.ffn_gate\.weight$=iq2_s
+^blk\.4\.ffn_up\.weight$=iq3_s
+^blk\.5\.attn_k\.weight$=iq4_xs
+^blk\.5\.attn_output\.weight$=iq4_xs
+^blk\.5\.attn_q\.weight$=iq2_s
+^blk\.5\.attn_v\.weight$=q5_k
+^blk\.5\.ffn_down\.weight$=iq3_xxs
+^blk\.5\.ffn_gate\.weight$=iq2_s
+^blk\.5\.ffn_up\.weight$=iq3_s
+^blk\.6\.attn_k\.weight$=iq4_xs
+^blk\.6\.attn_output\.weight$=iq4_xs
+^blk\.6\.attn_q\.weight$=iq2_s
+^blk\.6\.attn_v\.weight$=q5_k
+^blk\.6\.ffn_down\.weight$=iq3_xxs
+^blk\.6\.ffn_gate\.weight$=iq2_s
+^blk\.6\.ffn_up\.weight$=iq3_s
+^blk\.7\.attn_k\.weight$=iq4_xs
+^blk\.7\.attn_output\.weight$=iq4_xs
+^blk\.7\.attn_q\.weight$=iq2_s
+^blk\.7\.attn_v\.weight$=q5_k
+^blk\.7\.ffn_down\.weight$=iq3_xxs
+^blk\.7\.ffn_gate\.weight$=iq2_s
+^blk\.7\.ffn_up\.weight$=iq3_s
+^blk\.8\.attn_k\.weight$=iq4_xs
+^blk\.8\.attn_output\.weight$=iq4_xs
+^blk\.8\.attn_q\.weight$=iq2_s
+^blk\.8\.attn_v\.weight$=q5_k
+^blk\.8\.ffn_down\.weight$=iq3_xxs
+^blk\.8\.ffn_gate\.weight$=iq2_s
+^blk\.8\.ffn_up\.weight$=iq3_s
+^blk\.9\.attn_k\.weight$=iq4_xs
+^blk\.9\.attn_output\.weight$=iq4_xs
+^blk\.9\.attn_q\.weight$=iq2_s
+^blk\.9\.attn_v\.weight$=q5_k
+^blk\.9\.ffn_down\.weight$=iq3_xxs
+^blk\.9\.ffn_gate\.weight$=iq2_s
+^blk\.9\.ffn_up\.weight$=iq3_s
+^blk\.10\.attn_k\.weight$=iq4_xs
+^blk\.10\.attn_output\.weight$=iq4_xs
+^blk\.10\.attn_q\.weight$=iq2_s
+^blk\.10\.attn_v\.weight$=q5_k
+^blk\.10\.ffn_down\.weight$=iq2_s
+^blk\.10\.ffn_gate\.weight$=iq2_s
+^blk\.10\.ffn_up\.weight$=iq2_s
+^blk\.11\.attn_k\.weight$=iq4_xs
+^blk\.11\.attn_output\.weight$=iq4_xs
+^blk\.11\.attn_q\.weight$=iq2_s
+^blk\.11\.attn_v\.weight$=q5_k
+^blk\.11\.ffn_down\.weight$=iq2_s
+^blk\.11\.ffn_gate\.weight$=iq2_s
+^blk\.11\.ffn_up\.weight$=iq2_s
+^blk\.12\.attn_k\.weight$=iq4_xs
+^blk\.12\.attn_output\.weight$=iq4_xs
+^blk\.12\.attn_q\.weight$=iq2_s
+^blk\.12\.attn_v\.weight$=q5_k
+^blk\.12\.ffn_down\.weight$=iq2_s
+^blk\.12\.ffn_gate\.weight$=iq2_s
+^blk\.12\.ffn_up\.weight$=iq2_s
+^blk\.13\.attn_k\.weight$=iq4_xs
+^blk\.13\.attn_output\.weight$=iq4_xs
+^blk\.13\.attn_q\.weight$=iq2_s
+^blk\.13\.attn_v\.weight$=q5_k
+^blk\.13\.ffn_down\.weight$=iq2_s
+^blk\.13\.ffn_gate\.weight$=iq2_s
+^blk\.13\.ffn_up\.weight$=iq2_s
+^blk\.14\.attn_k\.weight$=iq4_xs
+^blk\.14\.attn_output\.weight$=iq4_xs
+^blk\.14\.attn_q\.weight$=iq2_s
+^blk\.14\.attn_v\.weight$=q5_k
+^blk\.14\.ffn_down\.weight$=iq2_s
+^blk\.14\.ffn_gate\.weight$=iq2_s
+^blk\.14\.ffn_up\.weight$=iq2_s
+^blk\.15\.attn_k\.weight$=iq4_xs
+^blk\.15\.attn_output\.weight$=iq4_xs
+^blk\.15\.attn_q\.weight$=iq2_s
+^blk\.15\.attn_v\.weight$=q5_k
+^blk\.15\.ffn_down\.weight$=iq2_s
+^blk\.15\.ffn_gate\.weight$=iq2_s
+^blk\.15\.ffn_up\.weight$=iq2_s
+^blk\.16\.attn_k\.weight$=iq4_xs
+^blk\.16\.attn_output\.weight$=iq4_xs
+^blk\.16\.attn_q\.weight$=iq2_s
+^blk\.16\.attn_v\.weight$=q5_k
+^blk\.16\.ffn_down\.weight$=iq2_s
+^blk\.16\.ffn_gate\.weight$=iq2_s
+^blk\.16\.ffn_up\.weight$=iq2_s
+^blk\.17\.attn_k\.weight$=iq4_xs
+^blk\.17\.attn_output\.weight$=iq4_xs
+^blk\.17\.attn_q\.weight$=iq2_s
+^blk\.17\.attn_v\.weight$=q5_k
+^blk\.17\.ffn_down\.weight$=iq2_s
+^blk\.17\.ffn_gate\.weight$=iq2_s
+^blk\.17\.ffn_up\.weight$=iq2_s
+^blk\.18\.attn_k\.weight$=iq4_xs
+^blk\.18\.attn_output\.weight$=iq4_xs
+^blk\.18\.attn_q\.weight$=iq2_s
+^blk\.18\.attn_v\.weight$=q5_k
+^blk\.18\.ffn_down\.weight$=iq2_s
+^blk\.18\.ffn_gate\.weight$=iq2_s
+^blk\.18\.ffn_up\.weight$=iq2_s
+^blk\.19\.attn_k\.weight$=iq4_xs
+^blk\.19\.attn_output\.weight$=iq4_xs
+^blk\.19\.attn_q\.weight$=iq2_s
+^blk\.19\.attn_v\.weight$=q5_k
+^blk\.19\.ffn_down\.weight$=iq2_s
+^blk\.19\.ffn_gate\.weight$=iq2_s
+^blk\.19\.ffn_up\.weight$=iq2_s
+^blk\.20\.attn_k\.weight$=iq4_xs
+^blk\.20\.attn_output\.weight$=iq4_xs
+^blk\.20\.attn_q\.weight$=iq2_s
+^blk\.20\.attn_v\.weight$=q5_k
+^blk\.20\.ffn_down\.weight$=iq2_s
+^blk\.20\.ffn_gate\.weight$=iq2_s
+^blk\.20\.ffn_up\.weight$=iq2_s
+^blk\.21\.attn_k\.weight$=iq4_xs
+^blk\.21\.attn_output\.weight$=iq4_xs
+^blk\.21\.attn_q\.weight$=iq2_s
+^blk\.21\.attn_v\.weight$=q5_k
+^blk\.21\.ffn_down\.weight$=iq2_s
+^blk\.21\.ffn_gate\.weight$=iq2_s
+^blk\.21\.ffn_up\.weight$=iq2_s
+^blk\.22\.attn_k\.weight$=iq4_xs
+^blk\.22\.attn_output\.weight$=iq4_xs
+^blk\.22\.attn_q\.weight$=iq2_s
+^blk\.22\.attn_v\.weight$=q5_k
+^blk\.22\.ffn_down\.weight$=iq2_s
+^blk\.22\.ffn_gate\.weight$=iq2_s
+^blk\.22\.ffn_up\.weight$=iq2_s
+^blk\.23\.attn_k\.weight$=iq4_xs
+^blk\.23\.attn_output\.weight$=iq4_xs
+^blk\.23\.attn_q\.weight$=iq2_s
+^blk\.23\.attn_v\.weight$=q5_k
+^blk\.23\.ffn_down\.weight$=iq2_s
+^blk\.23\.ffn_gate\.weight$=iq2_s
+^blk\.23\.ffn_up\.weight$=iq2_s
+^blk\.24\.attn_k\.weight$=iq4_xs
+^blk\.24\.attn_output\.weight$=iq4_xs
+^blk\.24\.attn_q\.weight$=iq2_s
+^blk\.24\.attn_v\.weight$=q5_k
+^blk\.24\.ffn_down\.weight$=iq2_s
+^blk\.24\.ffn_gate\.weight$=iq2_s
+^blk\.24\.ffn_up\.weight$=iq2_s
+^blk\.25\.attn_k\.weight$=iq4_xs
+^blk\.25\.attn_output\.weight$=iq4_xs
+^blk\.25\.attn_q\.weight$=iq2_s
+^blk\.25\.attn_v\.weight$=q5_k
+^blk\.25\.ffn_down\.weight$=iq2_s
+^blk\.25\.ffn_gate\.weight$=iq2_s
+^blk\.25\.ffn_up\.weight$=iq2_s
+^blk\.26\.attn_k\.weight$=iq4_xs
+^blk\.26\.attn_output\.weight$=iq4_xs
+^blk\.26\.attn_q\.weight$=iq2_s
+^blk\.26\.attn_v\.weight$=q5_k
+^blk\.26\.ffn_down\.weight$=iq2_s
+^blk\.26\.ffn_gate\.weight$=iq2_s
+^blk\.26\.ffn_up\.weight$=iq2_s
+^blk\.27\.attn_k\.weight$=iq4_xs
+^blk\.27\.attn_output\.weight$=iq4_xs
+^blk\.27\.attn_q\.weight$=iq2_s
+^blk\.27\.attn_v\.weight$=q5_k
+^blk\.27\.ffn_down\.weight$=iq2_s
+^blk\.27\.ffn_gate\.weight$=iq2_s
+^blk\.27\.ffn_up\.weight$=iq2_s
+^blk\.28\.attn_k\.weight$=iq4_xs
+^blk\.28\.attn_output\.weight$=iq4_xs
+^blk\.28\.attn_q\.weight$=iq2_s
+^blk\.28\.attn_v\.weight$=q5_k
+^blk\.28\.ffn_down\.weight$=iq2_s
+^blk\.28\.ffn_gate\.weight$=iq2_s
+^blk\.28\.ffn_up\.weight$=iq2_s
+^blk\.29\.attn_k\.weight$=iq4_xs
+^blk\.29\.attn_output\.weight$=iq4_xs
+^blk\.29\.attn_q\.weight$=iq2_s
+^blk\.29\.attn_v\.weight$=q5_k
+^blk\.29\.ffn_down\.weight$=iq2_s
+^blk\.29\.ffn_gate\.weight$=iq2_s
+^blk\.29\.ffn_up\.weight$=iq2_s
+^blk\.30\.attn_k\.weight$=iq4_xs
+^blk\.30\.attn_output\.weight$=iq2_s
+^blk\.30\.attn_q\.weight$=iq2_s
+^blk\.30\.attn_v\.weight$=q4_k
+^blk\.30\.ffn_down\.weight$=iq2_s
+^blk\.30\.ffn_gate\.weight$=iq2_s
+^blk\.30\.ffn_up\.weight$=iq2_s
+^blk\.31\.attn_k\.weight$=iq4_xs
+^blk\.31\.attn_output\.weight$=iq2_s
+^blk\.31\.attn_q\.weight$=iq2_s
+^blk\.31\.attn_v\.weight$=iq4_xs
+^blk\.31\.ffn_down\.weight$=iq2_s
+^blk\.31\.ffn_gate\.weight$=iq2_s
+^blk\.31\.ffn_up\.weight$=iq2_s
+^blk\.32\.attn_k\.weight$=iq4_xs
+^blk\.32\.attn_output\.weight$=iq2_s
+^blk\.32\.attn_q\.weight$=iq2_s
+^blk\.32\.attn_v\.weight$=iq4_xs
+^blk\.32\.ffn_down\.weight$=iq2_s
+^blk\.32\.ffn_gate\.weight$=iq2_s
+^blk\.32\.ffn_up\.weight$=iq2_s
+^blk\.33\.attn_k\.weight$=iq4_xs
+^blk\.33\.attn_output\.weight$=iq2_s
+^blk\.33\.attn_q\.weight$=iq2_s
+^blk\.33\.attn_v\.weight$=iq4_xs
+^blk\.33\.ffn_down\.weight$=iq2_s
+^blk\.33\.ffn_gate\.weight$=iq2_s
+^blk\.33\.ffn_up\.weight$=iq2_s
+^blk\.34\.attn_k\.weight$=iq4_xs
+^blk\.34\.attn_output\.weight$=iq2_s
+^blk\.34\.attn_q\.weight$=iq2_s
+^blk\.34\.attn_v\.weight$=iq4_xs
+^blk\.34\.ffn_down\.weight$=iq2_s
+^blk\.34\.ffn_gate\.weight$=iq2_s
+^blk\.34\.ffn_up\.weight$=iq2_s
+^blk\.35\.attn_k\.weight$=iq4_xs
+^blk\.35\.attn_output\.weight$=iq2_s
+^blk\.35\.attn_q\.weight$=iq2_s
+^blk\.35\.attn_v\.weight$=iq4_xs
+^blk\.35\.ffn_down\.weight$=iq2_s
+^blk\.35\.ffn_gate\.weight$=iq2_s
+^blk\.35\.ffn_up\.weight$=iq2_s
+^blk\.36\.attn_k\.weight$=iq4_xs
+^blk\.36\.attn_output\.weight$=iq2_s
+^blk\.36\.attn_q\.weight$=iq2_s
+^blk\.36\.attn_v\.weight$=iq4_xs
+^blk\.36\.ffn_down\.weight$=iq2_s
+^blk\.36\.ffn_gate\.weight$=iq2_s
+^blk\.36\.ffn_up\.weight$=iq2_s
+^blk\.37\.attn_k\.weight$=iq4_xs
+^blk\.37\.attn_output\.weight$=iq2_s
+^blk\.37\.attn_q\.weight$=iq2_s
+^blk\.37\.attn_v\.weight$=iq4_xs
+^blk\.37\.ffn_down\.weight$=iq2_s
+^blk\.37\.ffn_gate\.weight$=iq2_s
+^blk\.37\.ffn_up\.weight$=iq2_s
+^blk\.38\.attn_k\.weight$=iq4_xs
+^blk\.38\.attn_output\.weight$=iq2_s
+^blk\.38\.attn_q\.weight$=iq2_s
+^blk\.38\.attn_v\.weight$=iq4_xs
+^blk\.38\.ffn_down\.weight$=iq2_s
+^blk\.38\.ffn_gate\.weight$=iq2_s
+^blk\.38\.ffn_up\.weight$=iq2_s
+^blk\.39\.attn_k\.weight$=iq4_xs
+^blk\.39\.attn_output\.weight$=iq2_s
+^blk\.39\.attn_q\.weight$=iq2_s
+^blk\.39\.attn_v\.weight$=q5_k
+^blk\.39\.ffn_down\.weight$=iq3_s
+^blk\.39\.ffn_gate\.weight$=iq2_s
+^blk\.39\.ffn_up\.weight$=iq4_xs

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ2_S.layout.json

@@ -0,0 +1,88 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ2_S",
+  "base_type": "iq2_xs",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0959,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0599,
+      "type": "q3_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q3_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 5567554280,
+    "map_body+embd": 5089264360,
+    "ratio": 0.9141,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 2.556,
+  "body_bpw_heuristic": 2.846,
+  "total_params": 14765593600,
+  "file_bytes_pred": 5090964200,
+  "file_bpw_pred": 2.758,
+  "histogram": {
+    "attn_k": {
+      "iq4_xs": 30,
+      "iq3_s": 10
+    },
+    "attn_output": {
+      "iq3_s": 21,
+      "iq2_xs": 10,
+      "iq4_xs": 9
+    },
+    "attn_q": {
+      "iq2_xs": 40
+    },
+    "attn_v": {
+      "iq4_xs": 20,
+      "q4_k": 20
+    },
+    "ffn_down": {
+      "iq2_xs": 30,
+      "iq2_s": 8,
+      "iq3_xxs": 1,
+      "iq3_s": 1
+    },
+    "ffn_gate": {
+      "iq2_xs": 40
+    },
+    "ffn_up": {
+      "iq2_xs": 30,
+      "iq3_s": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ2_S",
+  "quantize_ftype": "IQ2_S",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ2_S.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q3_k
+^blk\.0\.attn_k\.weight$=iq4_xs
+^blk\.0\.attn_output\.weight$=iq3_s
+^blk\.0\.attn_q\.weight$=iq2_xs
+^blk\.0\.attn_v\.weight$=iq4_xs
+^blk\.0\.ffn_down\.weight$=iq2_xs
+^blk\.0\.ffn_gate\.weight$=iq2_xs
+^blk\.0\.ffn_up\.weight$=iq2_xs
+^blk\.1\.attn_k\.weight$=iq4_xs
+^blk\.1\.attn_output\.weight$=iq4_xs
+^blk\.1\.attn_q\.weight$=iq2_xs
+^blk\.1\.attn_v\.weight$=q4_k
+^blk\.1\.ffn_down\.weight$=iq3_xxs
+^blk\.1\.ffn_gate\.weight$=iq2_xs
+^blk\.1\.ffn_up\.weight$=iq3_s
+^blk\.2\.attn_k\.weight$=iq4_xs
+^blk\.2\.attn_output\.weight$=iq4_xs
+^blk\.2\.attn_q\.weight$=iq2_xs
+^blk\.2\.attn_v\.weight$=q4_k
+^blk\.2\.ffn_down\.weight$=iq2_s
+^blk\.2\.ffn_gate\.weight$=iq2_xs
+^blk\.2\.ffn_up\.weight$=iq3_s
+^blk\.3\.attn_k\.weight$=iq4_xs
+^blk\.3\.attn_output\.weight$=iq4_xs
+^blk\.3\.attn_q\.weight$=iq2_xs
+^blk\.3\.attn_v\.weight$=q4_k
+^blk\.3\.ffn_down\.weight$=iq2_s
+^blk\.3\.ffn_gate\.weight$=iq2_xs
+^blk\.3\.ffn_up\.weight$=iq3_s
+^blk\.4\.attn_k\.weight$=iq4_xs
+^blk\.4\.attn_output\.weight$=iq4_xs
+^blk\.4\.attn_q\.weight$=iq2_xs
+^blk\.4\.attn_v\.weight$=q4_k
+^blk\.4\.ffn_down\.weight$=iq2_s
+^blk\.4\.ffn_gate\.weight$=iq2_xs
+^blk\.4\.ffn_up\.weight$=iq3_s
+^blk\.5\.attn_k\.weight$=iq4_xs
+^blk\.5\.attn_output\.weight$=iq4_xs
+^blk\.5\.attn_q\.weight$=iq2_xs
+^blk\.5\.attn_v\.weight$=q4_k
+^blk\.5\.ffn_down\.weight$=iq2_s
+^blk\.5\.ffn_gate\.weight$=iq2_xs
+^blk\.5\.ffn_up\.weight$=iq3_s
+^blk\.6\.attn_k\.weight$=iq4_xs
+^blk\.6\.attn_output\.weight$=iq4_xs
+^blk\.6\.attn_q\.weight$=iq2_xs
+^blk\.6\.attn_v\.weight$=q4_k
+^blk\.6\.ffn_down\.weight$=iq2_s
+^blk\.6\.ffn_gate\.weight$=iq2_xs
+^blk\.6\.ffn_up\.weight$=iq3_s
+^blk\.7\.attn_k\.weight$=iq4_xs
+^blk\.7\.attn_output\.weight$=iq4_xs
+^blk\.7\.attn_q\.weight$=iq2_xs
+^blk\.7\.attn_v\.weight$=q4_k
+^blk\.7\.ffn_down\.weight$=iq2_s
+^blk\.7\.ffn_gate\.weight$=iq2_xs
+^blk\.7\.ffn_up\.weight$=iq3_s
+^blk\.8\.attn_k\.weight$=iq4_xs
+^blk\.8\.attn_output\.weight$=iq4_xs
+^blk\.8\.attn_q\.weight$=iq2_xs
+^blk\.8\.attn_v\.weight$=q4_k
+^blk\.8\.ffn_down\.weight$=iq2_s
+^blk\.8\.ffn_gate\.weight$=iq2_xs
+^blk\.8\.ffn_up\.weight$=iq3_s
+^blk\.9\.attn_k\.weight$=iq4_xs
+^blk\.9\.attn_output\.weight$=iq4_xs
+^blk\.9\.attn_q\.weight$=iq2_xs
+^blk\.9\.attn_v\.weight$=q4_k
+^blk\.9\.ffn_down\.weight$=iq2_s
+^blk\.9\.ffn_gate\.weight$=iq2_xs
+^blk\.9\.ffn_up\.weight$=iq3_s
+^blk\.10\.attn_k\.weight$=iq4_xs
+^blk\.10\.attn_output\.weight$=iq3_s
+^blk\.10\.attn_q\.weight$=iq2_xs
+^blk\.10\.attn_v\.weight$=q4_k
+^blk\.10\.ffn_down\.weight$=iq2_xs
+^blk\.10\.ffn_gate\.weight$=iq2_xs
+^blk\.10\.ffn_up\.weight$=iq2_xs
+^blk\.11\.attn_k\.weight$=iq4_xs
+^blk\.11\.attn_output\.weight$=iq3_s
+^blk\.11\.attn_q\.weight$=iq2_xs
+^blk\.11\.attn_v\.weight$=q4_k
+^blk\.11\.ffn_down\.weight$=iq2_xs
+^blk\.11\.ffn_gate\.weight$=iq2_xs
+^blk\.11\.ffn_up\.weight$=iq2_xs
+^blk\.12\.attn_k\.weight$=iq4_xs
+^blk\.12\.attn_output\.weight$=iq3_s
+^blk\.12\.attn_q\.weight$=iq2_xs
+^blk\.12\.attn_v\.weight$=q4_k
+^blk\.12\.ffn_down\.weight$=iq2_xs
+^blk\.12\.ffn_gate\.weight$=iq2_xs
+^blk\.12\.ffn_up\.weight$=iq2_xs
+^blk\.13\.attn_k\.weight$=iq4_xs
+^blk\.13\.attn_output\.weight$=iq3_s
+^blk\.13\.attn_q\.weight$=iq2_xs
+^blk\.13\.attn_v\.weight$=q4_k
+^blk\.13\.ffn_down\.weight$=iq2_xs
+^blk\.13\.ffn_gate\.weight$=iq2_xs
+^blk\.13\.ffn_up\.weight$=iq2_xs
+^blk\.14\.attn_k\.weight$=iq4_xs
+^blk\.14\.attn_output\.weight$=iq3_s
+^blk\.14\.attn_q\.weight$=iq2_xs
+^blk\.14\.attn_v\.weight$=q4_k
+^blk\.14\.ffn_down\.weight$=iq2_xs
+^blk\.14\.ffn_gate\.weight$=iq2_xs
+^blk\.14\.ffn_up\.weight$=iq2_xs
+^blk\.15\.attn_k\.weight$=iq4_xs
+^blk\.15\.attn_output\.weight$=iq3_s
+^blk\.15\.attn_q\.weight$=iq2_xs
+^blk\.15\.attn_v\.weight$=q4_k
+^blk\.15\.ffn_down\.weight$=iq2_xs
+^blk\.15\.ffn_gate\.weight$=iq2_xs
+^blk\.15\.ffn_up\.weight$=iq2_xs
+^blk\.16\.attn_k\.weight$=iq4_xs
+^blk\.16\.attn_output\.weight$=iq3_s
+^blk\.16\.attn_q\.weight$=iq2_xs
+^blk\.16\.attn_v\.weight$=q4_k
+^blk\.16\.ffn_down\.weight$=iq2_xs
+^blk\.16\.ffn_gate\.weight$=iq2_xs
+^blk\.16\.ffn_up\.weight$=iq2_xs
+^blk\.17\.attn_k\.weight$=iq4_xs
+^blk\.17\.attn_output\.weight$=iq3_s
+^blk\.17\.attn_q\.weight$=iq2_xs
+^blk\.17\.attn_v\.weight$=q4_k
+^blk\.17\.ffn_down\.weight$=iq2_xs
+^blk\.17\.ffn_gate\.weight$=iq2_xs
+^blk\.17\.ffn_up\.weight$=iq2_xs
+^blk\.18\.attn_k\.weight$=iq4_xs
+^blk\.18\.attn_output\.weight$=iq3_s
+^blk\.18\.attn_q\.weight$=iq2_xs
+^blk\.18\.attn_v\.weight$=q4_k
+^blk\.18\.ffn_down\.weight$=iq2_xs
+^blk\.18\.ffn_gate\.weight$=iq2_xs
+^blk\.18\.ffn_up\.weight$=iq2_xs
+^blk\.19\.attn_k\.weight$=iq4_xs
+^blk\.19\.attn_output\.weight$=iq3_s
+^blk\.19\.attn_q\.weight$=iq2_xs
+^blk\.19\.attn_v\.weight$=q4_k
+^blk\.19\.ffn_down\.weight$=iq2_xs
+^blk\.19\.ffn_gate\.weight$=iq2_xs
+^blk\.19\.ffn_up\.weight$=iq2_xs
+^blk\.20\.attn_k\.weight$=iq4_xs
+^blk\.20\.attn_output\.weight$=iq3_s
+^blk\.20\.attn_q\.weight$=iq2_xs
+^blk\.20\.attn_v\.weight$=q4_k
+^blk\.20\.ffn_down\.weight$=iq2_xs
+^blk\.20\.ffn_gate\.weight$=iq2_xs
+^blk\.20\.ffn_up\.weight$=iq2_xs
+^blk\.21\.attn_k\.weight$=iq4_xs
+^blk\.21\.attn_output\.weight$=iq3_s
+^blk\.21\.attn_q\.weight$=iq2_xs
+^blk\.21\.attn_v\.weight$=iq4_xs
+^blk\.21\.ffn_down\.weight$=iq2_xs
+^blk\.21\.ffn_gate\.weight$=iq2_xs
+^blk\.21\.ffn_up\.weight$=iq2_xs
+^blk\.22\.attn_k\.weight$=iq4_xs
+^blk\.22\.attn_output\.weight$=iq3_s
+^blk\.22\.attn_q\.weight$=iq2_xs
+^blk\.22\.attn_v\.weight$=iq4_xs
+^blk\.22\.ffn_down\.weight$=iq2_xs
+^blk\.22\.ffn_gate\.weight$=iq2_xs
+^blk\.22\.ffn_up\.weight$=iq2_xs
+^blk\.23\.attn_k\.weight$=iq4_xs
+^blk\.23\.attn_output\.weight$=iq3_s
+^blk\.23\.attn_q\.weight$=iq2_xs
+^blk\.23\.attn_v\.weight$=iq4_xs
+^blk\.23\.ffn_down\.weight$=iq2_xs
+^blk\.23\.ffn_gate\.weight$=iq2_xs
+^blk\.23\.ffn_up\.weight$=iq2_xs
+^blk\.24\.attn_k\.weight$=iq4_xs
+^blk\.24\.attn_output\.weight$=iq3_s
+^blk\.24\.attn_q\.weight$=iq2_xs
+^blk\.24\.attn_v\.weight$=iq4_xs
+^blk\.24\.ffn_down\.weight$=iq2_xs
+^blk\.24\.ffn_gate\.weight$=iq2_xs
+^blk\.24\.ffn_up\.weight$=iq2_xs
+^blk\.25\.attn_k\.weight$=iq4_xs
+^blk\.25\.attn_output\.weight$=iq3_s
+^blk\.25\.attn_q\.weight$=iq2_xs
+^blk\.25\.attn_v\.weight$=iq4_xs
+^blk\.25\.ffn_down\.weight$=iq2_xs
+^blk\.25\.ffn_gate\.weight$=iq2_xs
+^blk\.25\.ffn_up\.weight$=iq2_xs
+^blk\.26\.attn_k\.weight$=iq4_xs
+^blk\.26\.attn_output\.weight$=iq3_s
+^blk\.26\.attn_q\.weight$=iq2_xs
+^blk\.26\.attn_v\.weight$=iq4_xs
+^blk\.26\.ffn_down\.weight$=iq2_xs
+^blk\.26\.ffn_gate\.weight$=iq2_xs
+^blk\.26\.ffn_up\.weight$=iq2_xs
+^blk\.27\.attn_k\.weight$=iq4_xs
+^blk\.27\.attn_output\.weight$=iq3_s
+^blk\.27\.attn_q\.weight$=iq2_xs
+^blk\.27\.attn_v\.weight$=iq4_xs
+^blk\.27\.ffn_down\.weight$=iq2_xs
+^blk\.27\.ffn_gate\.weight$=iq2_xs
+^blk\.27\.ffn_up\.weight$=iq2_xs
+^blk\.28\.attn_k\.weight$=iq4_xs
+^blk\.28\.attn_output\.weight$=iq3_s
+^blk\.28\.attn_q\.weight$=iq2_xs
+^blk\.28\.attn_v\.weight$=iq4_xs
+^blk\.28\.ffn_down\.weight$=iq2_xs
+^blk\.28\.ffn_gate\.weight$=iq2_xs
+^blk\.28\.ffn_up\.weight$=iq2_xs
+^blk\.29\.attn_k\.weight$=iq4_xs
+^blk\.29\.attn_output\.weight$=iq3_s
+^blk\.29\.attn_q\.weight$=iq2_xs
+^blk\.29\.attn_v\.weight$=iq4_xs
+^blk\.29\.ffn_down\.weight$=iq2_xs
+^blk\.29\.ffn_gate\.weight$=iq2_xs
+^blk\.29\.ffn_up\.weight$=iq2_xs
+^blk\.30\.attn_k\.weight$=iq3_s
+^blk\.30\.attn_output\.weight$=iq2_xs
+^blk\.30\.attn_q\.weight$=iq2_xs
+^blk\.30\.attn_v\.weight$=iq4_xs
+^blk\.30\.ffn_down\.weight$=iq2_xs
+^blk\.30\.ffn_gate\.weight$=iq2_xs
+^blk\.30\.ffn_up\.weight$=iq2_xs
+^blk\.31\.attn_k\.weight$=iq3_s
+^blk\.31\.attn_output\.weight$=iq2_xs
+^blk\.31\.attn_q\.weight$=iq2_xs
+^blk\.31\.attn_v\.weight$=iq4_xs
+^blk\.31\.ffn_down\.weight$=iq2_xs
+^blk\.31\.ffn_gate\.weight$=iq2_xs
+^blk\.31\.ffn_up\.weight$=iq2_xs
+^blk\.32\.attn_k\.weight$=iq3_s
+^blk\.32\.attn_output\.weight$=iq2_xs
+^blk\.32\.attn_q\.weight$=iq2_xs
+^blk\.32\.attn_v\.weight$=iq4_xs
+^blk\.32\.ffn_down\.weight$=iq2_xs
+^blk\.32\.ffn_gate\.weight$=iq2_xs
+^blk\.32\.ffn_up\.weight$=iq2_xs
+^blk\.33\.attn_k\.weight$=iq3_s
+^blk\.33\.attn_output\.weight$=iq2_xs
+^blk\.33\.attn_q\.weight$=iq2_xs
+^blk\.33\.attn_v\.weight$=iq4_xs
+^blk\.33\.ffn_down\.weight$=iq2_xs
+^blk\.33\.ffn_gate\.weight$=iq2_xs
+^blk\.33\.ffn_up\.weight$=iq2_xs
+^blk\.34\.attn_k\.weight$=iq3_s
+^blk\.34\.attn_output\.weight$=iq2_xs
+^blk\.34\.attn_q\.weight$=iq2_xs
+^blk\.34\.attn_v\.weight$=iq4_xs
+^blk\.34\.ffn_down\.weight$=iq2_xs
+^blk\.34\.ffn_gate\.weight$=iq2_xs
+^blk\.34\.ffn_up\.weight$=iq2_xs
+^blk\.35\.attn_k\.weight$=iq3_s
+^blk\.35\.attn_output\.weight$=iq2_xs
+^blk\.35\.attn_q\.weight$=iq2_xs
+^blk\.35\.attn_v\.weight$=iq4_xs
+^blk\.35\.ffn_down\.weight$=iq2_xs
+^blk\.35\.ffn_gate\.weight$=iq2_xs
+^blk\.35\.ffn_up\.weight$=iq2_xs
+^blk\.36\.attn_k\.weight$=iq3_s
+^blk\.36\.attn_output\.weight$=iq2_xs
+^blk\.36\.attn_q\.weight$=iq2_xs
+^blk\.36\.attn_v\.weight$=iq4_xs
+^blk\.36\.ffn_down\.weight$=iq2_xs
+^blk\.36\.ffn_gate\.weight$=iq2_xs
+^blk\.36\.ffn_up\.weight$=iq2_xs
+^blk\.37\.attn_k\.weight$=iq3_s
+^blk\.37\.attn_output\.weight$=iq2_xs
+^blk\.37\.attn_q\.weight$=iq2_xs
+^blk\.37\.attn_v\.weight$=iq4_xs
+^blk\.37\.ffn_down\.weight$=iq2_xs
+^blk\.37\.ffn_gate\.weight$=iq2_xs
+^blk\.37\.ffn_up\.weight$=iq2_xs
+^blk\.38\.attn_k\.weight$=iq3_s
+^blk\.38\.attn_output\.weight$=iq2_xs
+^blk\.38\.attn_q\.weight$=iq2_xs
+^blk\.38\.attn_v\.weight$=iq4_xs
+^blk\.38\.ffn_down\.weight$=iq2_xs
+^blk\.38\.ffn_gate\.weight$=iq2_xs
+^blk\.38\.ffn_up\.weight$=iq2_xs
+^blk\.39\.attn_k\.weight$=iq3_s
+^blk\.39\.attn_output\.weight$=iq2_xs
+^blk\.39\.attn_q\.weight$=iq2_xs
+^blk\.39\.attn_v\.weight$=iq4_xs
+^blk\.39\.ffn_down\.weight$=iq3_s
+^blk\.39\.ffn_gate\.weight$=iq2_xs
+^blk\.39\.ffn_up\.weight$=iq3_s

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_M.layout.json

@@ -0,0 +1,87 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ3_M",
+  "base_type": "iq3_s",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.50",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "L",
+  "rung_share": 0.5,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0896,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.047,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 7105068800,
+    "map_body+embd": 8003526400,
+    "ratio": 1.1265,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.5,
+  "body_bpw": 4.258,
+  "body_bpw_heuristic": 3.714,
+  "total_params": 14765593600,
+  "file_bytes_pred": 8005226240,
+  "file_bpw_pred": 4.337,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 38,
+      "q6_k": 2
+    },
+    "attn_output": {
+      "q5_k": 31,
+      "iq3_s": 9
+    },
+    "attn_q": {
+      "iq3_s": 38,
+      "q4_k": 2
+    },
+    "attn_v": {
+      "q8_0": 29,
+      "q6_k": 11
+    },
+    "ffn_down": {
+      "q5_k": 20,
+      "iq3_s": 20
+    },
+    "ffn_gate": {
+      "iq3_s": 38,
+      "q5_k": 2
+    },
+    "ffn_up": {
+      "q5_k": 20,
+      "iq3_s": 20
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ3_M",
+  "quantize_ftype": "IQ3_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_M.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q6_k
+^blk\.0\.attn_output\.weight$=q5_k
+^blk\.0\.attn_q\.weight$=iq3_s
+^blk\.0\.attn_v\.weight$=q6_k
+^blk\.0\.ffn_down\.weight$=q5_k
+^blk\.0\.ffn_gate\.weight$=iq3_s
+^blk\.0\.ffn_up\.weight$=q5_k
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=iq3_s
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q5_k
+^blk\.1\.ffn_gate\.weight$=q5_k
+^blk\.1\.ffn_up\.weight$=q5_k
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=iq3_s
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q5_k
+^blk\.2\.ffn_gate\.weight$=iq3_s
+^blk\.2\.ffn_up\.weight$=q5_k
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=iq3_s
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q5_k
+^blk\.3\.ffn_gate\.weight$=iq3_s
+^blk\.3\.ffn_up\.weight$=q5_k
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=iq3_s
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q5_k
+^blk\.4\.ffn_gate\.weight$=iq3_s
+^blk\.4\.ffn_up\.weight$=q5_k
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=iq3_s
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q5_k
+^blk\.5\.ffn_gate\.weight$=iq3_s
+^blk\.5\.ffn_up\.weight$=q5_k
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=iq3_s
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q5_k
+^blk\.6\.ffn_gate\.weight$=iq3_s
+^blk\.6\.ffn_up\.weight$=q5_k
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=iq3_s
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q5_k
+^blk\.7\.ffn_gate\.weight$=iq3_s
+^blk\.7\.ffn_up\.weight$=q5_k
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=iq3_s
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q5_k
+^blk\.8\.ffn_gate\.weight$=iq3_s
+^blk\.8\.ffn_up\.weight$=q5_k
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=iq3_s
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q5_k
+^blk\.9\.ffn_gate\.weight$=iq3_s
+^blk\.9\.ffn_up\.weight$=q5_k
+^blk\.10\.attn_k\.weight$=q6_k
+^blk\.10\.attn_output\.weight$=q5_k
+^blk\.10\.attn_q\.weight$=q4_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=iq3_s
+^blk\.10\.ffn_gate\.weight$=iq3_s
+^blk\.10\.ffn_up\.weight$=iq3_s
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=q5_k
+^blk\.11\.attn_q\.weight$=q4_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=iq3_s
+^blk\.11\.ffn_gate\.weight$=iq3_s
+^blk\.11\.ffn_up\.weight$=iq3_s
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=q5_k
+^blk\.12\.attn_q\.weight$=iq3_s
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=iq3_s
+^blk\.12\.ffn_gate\.weight$=iq3_s
+^blk\.12\.ffn_up\.weight$=iq3_s
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=q5_k
+^blk\.13\.attn_q\.weight$=iq3_s
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=iq3_s
+^blk\.13\.ffn_gate\.weight$=iq3_s
+^blk\.13\.ffn_up\.weight$=iq3_s
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=q5_k
+^blk\.14\.attn_q\.weight$=iq3_s
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=iq3_s
+^blk\.14\.ffn_gate\.weight$=iq3_s
+^blk\.14\.ffn_up\.weight$=iq3_s
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=q5_k
+^blk\.15\.attn_q\.weight$=iq3_s
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=iq3_s
+^blk\.15\.ffn_gate\.weight$=iq3_s
+^blk\.15\.ffn_up\.weight$=iq3_s
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=q5_k
+^blk\.16\.attn_q\.weight$=iq3_s
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=iq3_s
+^blk\.16\.ffn_gate\.weight$=iq3_s
+^blk\.16\.ffn_up\.weight$=iq3_s
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=q5_k
+^blk\.17\.attn_q\.weight$=iq3_s
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=iq3_s
+^blk\.17\.ffn_gate\.weight$=iq3_s
+^blk\.17\.ffn_up\.weight$=iq3_s
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=q5_k
+^blk\.18\.attn_q\.weight$=iq3_s
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=iq3_s
+^blk\.18\.ffn_gate\.weight$=iq3_s
+^blk\.18\.ffn_up\.weight$=iq3_s
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=q5_k
+^blk\.19\.attn_q\.weight$=iq3_s
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=iq3_s
+^blk\.19\.ffn_gate\.weight$=iq3_s
+^blk\.19\.ffn_up\.weight$=iq3_s
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=q5_k
+^blk\.20\.attn_q\.weight$=iq3_s
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=iq3_s
+^blk\.20\.ffn_gate\.weight$=iq3_s
+^blk\.20\.ffn_up\.weight$=iq3_s
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=q5_k
+^blk\.21\.attn_q\.weight$=iq3_s
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=iq3_s
+^blk\.21\.ffn_gate\.weight$=iq3_s
+^blk\.21\.ffn_up\.weight$=iq3_s
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=q5_k
+^blk\.22\.attn_q\.weight$=iq3_s
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=iq3_s
+^blk\.22\.ffn_gate\.weight$=iq3_s
+^blk\.22\.ffn_up\.weight$=iq3_s
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=q5_k
+^blk\.23\.attn_q\.weight$=iq3_s
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=iq3_s
+^blk\.23\.ffn_gate\.weight$=iq3_s
+^blk\.23\.ffn_up\.weight$=iq3_s
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=q5_k
+^blk\.24\.attn_q\.weight$=iq3_s
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=iq3_s
+^blk\.24\.ffn_gate\.weight$=iq3_s
+^blk\.24\.ffn_up\.weight$=iq3_s
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=q5_k
+^blk\.25\.attn_q\.weight$=iq3_s
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=iq3_s
+^blk\.25\.ffn_gate\.weight$=iq3_s
+^blk\.25\.ffn_up\.weight$=iq3_s
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=q5_k
+^blk\.26\.attn_q\.weight$=iq3_s
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=iq3_s
+^blk\.26\.ffn_gate\.weight$=iq3_s
+^blk\.26\.ffn_up\.weight$=iq3_s
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=q5_k
+^blk\.27\.attn_q\.weight$=iq3_s
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=iq3_s
+^blk\.27\.ffn_gate\.weight$=iq3_s
+^blk\.27\.ffn_up\.weight$=iq3_s
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=q5_k
+^blk\.28\.attn_q\.weight$=iq3_s
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=iq3_s
+^blk\.28\.ffn_gate\.weight$=iq3_s
+^blk\.28\.ffn_up\.weight$=iq3_s
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=q5_k
+^blk\.29\.attn_q\.weight$=iq3_s
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=iq3_s
+^blk\.29\.ffn_gate\.weight$=iq3_s
+^blk\.29\.ffn_up\.weight$=iq3_s
+^blk\.30\.attn_k\.weight$=q5_k
+^blk\.30\.attn_output\.weight$=iq3_s
+^blk\.30\.attn_q\.weight$=iq3_s
+^blk\.30\.attn_v\.weight$=q6_k
+^blk\.30\.ffn_down\.weight$=q5_k
+^blk\.30\.ffn_gate\.weight$=iq3_s
+^blk\.30\.ffn_up\.weight$=q5_k
+^blk\.31\.attn_k\.weight$=q5_k
+^blk\.31\.attn_output\.weight$=iq3_s
+^blk\.31\.attn_q\.weight$=iq3_s
+^blk\.31\.attn_v\.weight$=q6_k
+^blk\.31\.ffn_down\.weight$=q5_k
+^blk\.31\.ffn_gate\.weight$=iq3_s
+^blk\.31\.ffn_up\.weight$=q5_k
+^blk\.32\.attn_k\.weight$=q5_k
+^blk\.32\.attn_output\.weight$=iq3_s
+^blk\.32\.attn_q\.weight$=iq3_s
+^blk\.32\.attn_v\.weight$=q6_k
+^blk\.32\.ffn_down\.weight$=q5_k
+^blk\.32\.ffn_gate\.weight$=iq3_s
+^blk\.32\.ffn_up\.weight$=q5_k
+^blk\.33\.attn_k\.weight$=q5_k
+^blk\.33\.attn_output\.weight$=iq3_s
+^blk\.33\.attn_q\.weight$=iq3_s
+^blk\.33\.attn_v\.weight$=q6_k
+^blk\.33\.ffn_down\.weight$=q5_k
+^blk\.33\.ffn_gate\.weight$=iq3_s
+^blk\.33\.ffn_up\.weight$=q5_k
+^blk\.34\.attn_k\.weight$=q5_k
+^blk\.34\.attn_output\.weight$=iq3_s
+^blk\.34\.attn_q\.weight$=iq3_s
+^blk\.34\.attn_v\.weight$=q6_k
+^blk\.34\.ffn_down\.weight$=q5_k
+^blk\.34\.ffn_gate\.weight$=iq3_s
+^blk\.34\.ffn_up\.weight$=q5_k
+^blk\.35\.attn_k\.weight$=q5_k
+^blk\.35\.attn_output\.weight$=iq3_s
+^blk\.35\.attn_q\.weight$=iq3_s
+^blk\.35\.attn_v\.weight$=q6_k
+^blk\.35\.ffn_down\.weight$=q5_k
+^blk\.35\.ffn_gate\.weight$=iq3_s
+^blk\.35\.ffn_up\.weight$=q5_k
+^blk\.36\.attn_k\.weight$=q5_k
+^blk\.36\.attn_output\.weight$=iq3_s
+^blk\.36\.attn_q\.weight$=iq3_s
+^blk\.36\.attn_v\.weight$=q6_k
+^blk\.36\.ffn_down\.weight$=q5_k
+^blk\.36\.ffn_gate\.weight$=iq3_s
+^blk\.36\.ffn_up\.weight$=q5_k
+^blk\.37\.attn_k\.weight$=q5_k
+^blk\.37\.attn_output\.weight$=iq3_s
+^blk\.37\.attn_q\.weight$=iq3_s
+^blk\.37\.attn_v\.weight$=q6_k
+^blk\.37\.ffn_down\.weight$=q5_k
+^blk\.37\.ffn_gate\.weight$=iq3_s
+^blk\.37\.ffn_up\.weight$=q5_k
+^blk\.38\.attn_k\.weight$=q5_k
+^blk\.38\.attn_output\.weight$=iq3_s
+^blk\.38\.attn_q\.weight$=iq3_s
+^blk\.38\.attn_v\.weight$=q6_k
+^blk\.38\.ffn_down\.weight$=q5_k
+^blk\.38\.ffn_gate\.weight$=iq3_s
+^blk\.38\.ffn_up\.weight$=q5_k
+^blk\.39\.attn_k\.weight$=q5_k
+^blk\.39\.attn_output\.weight$=q5_k
+^blk\.39\.attn_q\.weight$=iq3_s
+^blk\.39\.attn_v\.weight$=q6_k
+^blk\.39\.ffn_down\.weight$=q5_k
+^blk\.39\.ffn_gate\.weight$=q5_k
+^blk\.39\.ffn_up\.weight$=q5_k

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_XS.layout.json

@@ -0,0 +1,85 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ3_XS",
+  "base_type": "iq3_s",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0961,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0503,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 6629728000,
+    "map_body+embd": 6872743680,
+    "ratio": 1.0367,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.9,
+  "body_bpw": 3.574,
+  "body_bpw_heuristic": 3.427,
+  "total_params": 14765593600,
+  "file_bytes_pred": 6874443520,
+  "file_bpw_pred": 3.725,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 40
+    },
+    "attn_output": {
+      "iq3_s": 27,
+      "q5_k": 9,
+      "iq4_xs": 3,
+      "q4_k": 1
+    },
+    "attn_q": {
+      "iq3_s": 40
+    },
+    "attn_v": {
+      "q6_k": 31,
+      "q5_k": 9
+    },
+    "ffn_down": {
+      "iq3_s": 40
+    },
+    "ffn_gate": {
+      "iq3_s": 40
+    },
+    "ffn_up": {
+      "iq3_s": 39,
+      "q5_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ3_XS",
+  "quantize_ftype": "IQ3_XS",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_XS.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q5_k
+^blk\.0\.attn_output\.weight$=iq3_s
+^blk\.0\.attn_q\.weight$=iq3_s
+^blk\.0\.attn_v\.weight$=q6_k
+^blk\.0\.ffn_down\.weight$=iq3_s
+^blk\.0\.ffn_gate\.weight$=iq3_s
+^blk\.0\.ffn_up\.weight$=iq3_s
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=iq3_s
+^blk\.1\.attn_v\.weight$=q6_k
+^blk\.1\.ffn_down\.weight$=iq3_s
+^blk\.1\.ffn_gate\.weight$=iq3_s
+^blk\.1\.ffn_up\.weight$=iq3_s
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=iq3_s
+^blk\.2\.attn_v\.weight$=q6_k
+^blk\.2\.ffn_down\.weight$=iq3_s
+^blk\.2\.ffn_gate\.weight$=iq3_s
+^blk\.2\.ffn_up\.weight$=iq3_s
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=iq3_s
+^blk\.3\.attn_v\.weight$=q6_k
+^blk\.3\.ffn_down\.weight$=iq3_s
+^blk\.3\.ffn_gate\.weight$=iq3_s
+^blk\.3\.ffn_up\.weight$=iq3_s
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=iq3_s
+^blk\.4\.attn_v\.weight$=q6_k
+^blk\.4\.ffn_down\.weight$=iq3_s
+^blk\.4\.ffn_gate\.weight$=iq3_s
+^blk\.4\.ffn_up\.weight$=iq3_s
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=iq3_s
+^blk\.5\.attn_v\.weight$=q6_k
+^blk\.5\.ffn_down\.weight$=iq3_s
+^blk\.5\.ffn_gate\.weight$=iq3_s
+^blk\.5\.ffn_up\.weight$=iq3_s
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=iq3_s
+^blk\.6\.attn_v\.weight$=q6_k
+^blk\.6\.ffn_down\.weight$=iq3_s
+^blk\.6\.ffn_gate\.weight$=iq3_s
+^blk\.6\.ffn_up\.weight$=iq3_s
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=iq3_s
+^blk\.7\.attn_v\.weight$=q6_k
+^blk\.7\.ffn_down\.weight$=iq3_s
+^blk\.7\.ffn_gate\.weight$=iq3_s
+^blk\.7\.ffn_up\.weight$=iq3_s
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=iq3_s
+^blk\.8\.attn_v\.weight$=q6_k
+^blk\.8\.ffn_down\.weight$=iq3_s
+^blk\.8\.ffn_gate\.weight$=iq3_s
+^blk\.8\.ffn_up\.weight$=iq3_s
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=iq3_s
+^blk\.9\.attn_v\.weight$=q6_k
+^blk\.9\.ffn_down\.weight$=iq3_s
+^blk\.9\.ffn_gate\.weight$=iq3_s
+^blk\.9\.ffn_up\.weight$=iq3_s
+^blk\.10\.attn_k\.weight$=q5_k
+^blk\.10\.attn_output\.weight$=q4_k
+^blk\.10\.attn_q\.weight$=iq3_s
+^blk\.10\.attn_v\.weight$=q6_k
+^blk\.10\.ffn_down\.weight$=iq3_s
+^blk\.10\.ffn_gate\.weight$=iq3_s
+^blk\.10\.ffn_up\.weight$=iq3_s
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=iq4_xs
+^blk\.11\.attn_q\.weight$=iq3_s
+^blk\.11\.attn_v\.weight$=q6_k
+^blk\.11\.ffn_down\.weight$=iq3_s
+^blk\.11\.ffn_gate\.weight$=iq3_s
+^blk\.11\.ffn_up\.weight$=iq3_s
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=iq4_xs
+^blk\.12\.attn_q\.weight$=iq3_s
+^blk\.12\.attn_v\.weight$=q6_k
+^blk\.12\.ffn_down\.weight$=iq3_s
+^blk\.12\.ffn_gate\.weight$=iq3_s
+^blk\.12\.ffn_up\.weight$=iq3_s
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=iq4_xs
+^blk\.13\.attn_q\.weight$=iq3_s
+^blk\.13\.attn_v\.weight$=q6_k
+^blk\.13\.ffn_down\.weight$=iq3_s
+^blk\.13\.ffn_gate\.weight$=iq3_s
+^blk\.13\.ffn_up\.weight$=iq3_s
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=iq3_s
+^blk\.14\.attn_q\.weight$=iq3_s
+^blk\.14\.attn_v\.weight$=q6_k
+^blk\.14\.ffn_down\.weight$=iq3_s
+^blk\.14\.ffn_gate\.weight$=iq3_s
+^blk\.14\.ffn_up\.weight$=iq3_s
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=iq3_s
+^blk\.15\.attn_q\.weight$=iq3_s
+^blk\.15\.attn_v\.weight$=q6_k
+^blk\.15\.ffn_down\.weight$=iq3_s
+^blk\.15\.ffn_gate\.weight$=iq3_s
+^blk\.15\.ffn_up\.weight$=iq3_s
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=iq3_s
+^blk\.16\.attn_q\.weight$=iq3_s
+^blk\.16\.attn_v\.weight$=q6_k
+^blk\.16\.ffn_down\.weight$=iq3_s
+^blk\.16\.ffn_gate\.weight$=iq3_s
+^blk\.16\.ffn_up\.weight$=iq3_s
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=iq3_s
+^blk\.17\.attn_q\.weight$=iq3_s
+^blk\.17\.attn_v\.weight$=q6_k
+^blk\.17\.ffn_down\.weight$=iq3_s
+^blk\.17\.ffn_gate\.weight$=iq3_s
+^blk\.17\.ffn_up\.weight$=iq3_s
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=iq3_s
+^blk\.18\.attn_q\.weight$=iq3_s
+^blk\.18\.attn_v\.weight$=q6_k
+^blk\.18\.ffn_down\.weight$=iq3_s
+^blk\.18\.ffn_gate\.weight$=iq3_s
+^blk\.18\.ffn_up\.weight$=iq3_s
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=iq3_s
+^blk\.19\.attn_q\.weight$=iq3_s
+^blk\.19\.attn_v\.weight$=q6_k
+^blk\.19\.ffn_down\.weight$=iq3_s
+^blk\.19\.ffn_gate\.weight$=iq3_s
+^blk\.19\.ffn_up\.weight$=iq3_s
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=iq3_s
+^blk\.20\.attn_q\.weight$=iq3_s
+^blk\.20\.attn_v\.weight$=q6_k
+^blk\.20\.ffn_down\.weight$=iq3_s
+^blk\.20\.ffn_gate\.weight$=iq3_s
+^blk\.20\.ffn_up\.weight$=iq3_s
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=iq3_s
+^blk\.21\.attn_q\.weight$=iq3_s
+^blk\.21\.attn_v\.weight$=q6_k
+^blk\.21\.ffn_down\.weight$=iq3_s
+^blk\.21\.ffn_gate\.weight$=iq3_s
+^blk\.21\.ffn_up\.weight$=iq3_s
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=iq3_s
+^blk\.22\.attn_q\.weight$=iq3_s
+^blk\.22\.attn_v\.weight$=q6_k
+^blk\.22\.ffn_down\.weight$=iq3_s
+^blk\.22\.ffn_gate\.weight$=iq3_s
+^blk\.22\.ffn_up\.weight$=iq3_s
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=iq3_s
+^blk\.23\.attn_q\.weight$=iq3_s
+^blk\.23\.attn_v\.weight$=q6_k
+^blk\.23\.ffn_down\.weight$=iq3_s
+^blk\.23\.ffn_gate\.weight$=iq3_s
+^blk\.23\.ffn_up\.weight$=iq3_s
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=iq3_s
+^blk\.24\.attn_q\.weight$=iq3_s
+^blk\.24\.attn_v\.weight$=q6_k
+^blk\.24\.ffn_down\.weight$=iq3_s
+^blk\.24\.ffn_gate\.weight$=iq3_s
+^blk\.24\.ffn_up\.weight$=iq3_s
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=iq3_s
+^blk\.25\.attn_q\.weight$=iq3_s
+^blk\.25\.attn_v\.weight$=q6_k
+^blk\.25\.ffn_down\.weight$=iq3_s
+^blk\.25\.ffn_gate\.weight$=iq3_s
+^blk\.25\.ffn_up\.weight$=iq3_s
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=iq3_s
+^blk\.26\.attn_q\.weight$=iq3_s
+^blk\.26\.attn_v\.weight$=q6_k
+^blk\.26\.ffn_down\.weight$=iq3_s
+^blk\.26\.ffn_gate\.weight$=iq3_s
+^blk\.26\.ffn_up\.weight$=iq3_s
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=iq3_s
+^blk\.27\.attn_q\.weight$=iq3_s
+^blk\.27\.attn_v\.weight$=q6_k
+^blk\.27\.ffn_down\.weight$=iq3_s
+^blk\.27\.ffn_gate\.weight$=iq3_s
+^blk\.27\.ffn_up\.weight$=iq3_s
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=iq3_s
+^blk\.28\.attn_q\.weight$=iq3_s
+^blk\.28\.attn_v\.weight$=q6_k
+^blk\.28\.ffn_down\.weight$=iq3_s
+^blk\.28\.ffn_gate\.weight$=iq3_s
+^blk\.28\.ffn_up\.weight$=iq3_s
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=iq3_s
+^blk\.29\.attn_q\.weight$=iq3_s
+^blk\.29\.attn_v\.weight$=q6_k
+^blk\.29\.ffn_down\.weight$=iq3_s
+^blk\.29\.ffn_gate\.weight$=iq3_s
+^blk\.29\.ffn_up\.weight$=iq3_s
+^blk\.30\.attn_k\.weight$=q5_k
+^blk\.30\.attn_output\.weight$=iq3_s
+^blk\.30\.attn_q\.weight$=iq3_s
+^blk\.30\.attn_v\.weight$=q5_k
+^blk\.30\.ffn_down\.weight$=iq3_s
+^blk\.30\.ffn_gate\.weight$=iq3_s
+^blk\.30\.ffn_up\.weight$=iq3_s
+^blk\.31\.attn_k\.weight$=q5_k
+^blk\.31\.attn_output\.weight$=iq3_s
+^blk\.31\.attn_q\.weight$=iq3_s
+^blk\.31\.attn_v\.weight$=q5_k
+^blk\.31\.ffn_down\.weight$=iq3_s
+^blk\.31\.ffn_gate\.weight$=iq3_s
+^blk\.31\.ffn_up\.weight$=iq3_s
+^blk\.32\.attn_k\.weight$=q5_k
+^blk\.32\.attn_output\.weight$=iq3_s
+^blk\.32\.attn_q\.weight$=iq3_s
+^blk\.32\.attn_v\.weight$=q5_k
+^blk\.32\.ffn_down\.weight$=iq3_s
+^blk\.32\.ffn_gate\.weight$=iq3_s
+^blk\.32\.ffn_up\.weight$=iq3_s
+^blk\.33\.attn_k\.weight$=q5_k
+^blk\.33\.attn_output\.weight$=iq3_s
+^blk\.33\.attn_q\.weight$=iq3_s
+^blk\.33\.attn_v\.weight$=q5_k
+^blk\.33\.ffn_down\.weight$=iq3_s
+^blk\.33\.ffn_gate\.weight$=iq3_s
+^blk\.33\.ffn_up\.weight$=iq3_s
+^blk\.34\.attn_k\.weight$=q5_k
+^blk\.34\.attn_output\.weight$=iq3_s
+^blk\.34\.attn_q\.weight$=iq3_s
+^blk\.34\.attn_v\.weight$=q5_k
+^blk\.34\.ffn_down\.weight$=iq3_s
+^blk\.34\.ffn_gate\.weight$=iq3_s
+^blk\.34\.ffn_up\.weight$=iq3_s
+^blk\.35\.attn_k\.weight$=q5_k
+^blk\.35\.attn_output\.weight$=iq3_s
+^blk\.35\.attn_q\.weight$=iq3_s
+^blk\.35\.attn_v\.weight$=q5_k
+^blk\.35\.ffn_down\.weight$=iq3_s
+^blk\.35\.ffn_gate\.weight$=iq3_s
+^blk\.35\.ffn_up\.weight$=iq3_s
+^blk\.36\.attn_k\.weight$=q5_k
+^blk\.36\.attn_output\.weight$=iq3_s
+^blk\.36\.attn_q\.weight$=iq3_s
+^blk\.36\.attn_v\.weight$=q5_k
+^blk\.36\.ffn_down\.weight$=iq3_s
+^blk\.36\.ffn_gate\.weight$=iq3_s
+^blk\.36\.ffn_up\.weight$=iq3_s
+^blk\.37\.attn_k\.weight$=q5_k
+^blk\.37\.attn_output\.weight$=iq3_s
+^blk\.37\.attn_q\.weight$=iq3_s
+^blk\.37\.attn_v\.weight$=q5_k
+^blk\.37\.ffn_down\.weight$=iq3_s
+^blk\.37\.ffn_gate\.weight$=iq3_s
+^blk\.37\.ffn_up\.weight$=iq3_s
+^blk\.38\.attn_k\.weight$=q5_k
+^blk\.38\.attn_output\.weight$=iq3_s
+^blk\.38\.attn_q\.weight$=iq3_s
+^blk\.38\.attn_v\.weight$=q5_k
+^blk\.38\.ffn_down\.weight$=iq3_s
+^blk\.38\.ffn_gate\.weight$=iq3_s
+^blk\.38\.ffn_up\.weight$=iq3_s
+^blk\.39\.attn_k\.weight$=q5_k
+^blk\.39\.attn_output\.weight$=iq3_s
+^blk\.39\.attn_q\.weight$=iq3_s
+^blk\.39\.attn_v\.weight$=q6_k
+^blk\.39\.ffn_down\.weight$=iq3_s
+^blk\.39\.ffn_gate\.weight$=iq3_s
+^blk\.39\.ffn_up\.weight$=q5_k

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_XXS.layout.json

@@ -0,0 +1,87 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ3_XXS",
+  "base_type": "iq3_xxs",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0855,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0534,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 6241510120,
+    "map_body+embd": 6575824640,
+    "ratio": 1.0536,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 3.394,
+  "body_bpw_heuristic": 3.254,
+  "total_params": 14765593600,
+  "file_bytes_pred": 6577524480,
+  "file_bpw_pred": 3.564,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 30,
+      "q4_k": 10
+    },
+    "attn_output": {
+      "iq4_xs": 20,
+      "iq3_xxs": 11,
+      "q5_k": 9
+    },
+    "attn_q": {
+      "iq3_xxs": 40
+    },
+    "attn_v": {
+      "q5_k": 27,
+      "q6_k": 13
+    },
+    "ffn_down": {
+      "iq3_xxs": 30,
+      "iq4_xs": 10
+    },
+    "ffn_gate": {
+      "iq3_xxs": 40
+    },
+    "ffn_up": {
+      "iq3_xxs": 30,
+      "iq4_xs": 9,
+      "q4_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ3_XXS",
+  "quantize_ftype": "IQ3_XXS",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ3_XXS.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q5_k
+^blk\.0\.attn_output\.weight$=iq3_xxs
+^blk\.0\.attn_q\.weight$=iq3_xxs
+^blk\.0\.attn_v\.weight$=q5_k
+^blk\.0\.ffn_down\.weight$=iq3_xxs
+^blk\.0\.ffn_gate\.weight$=iq3_xxs
+^blk\.0\.ffn_up\.weight$=iq3_xxs
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=iq3_xxs
+^blk\.1\.attn_v\.weight$=q6_k
+^blk\.1\.ffn_down\.weight$=iq4_xs
+^blk\.1\.ffn_gate\.weight$=iq3_xxs
+^blk\.1\.ffn_up\.weight$=iq4_xs
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=iq3_xxs
+^blk\.2\.attn_v\.weight$=q6_k
+^blk\.2\.ffn_down\.weight$=iq4_xs
+^blk\.2\.ffn_gate\.weight$=iq3_xxs
+^blk\.2\.ffn_up\.weight$=iq4_xs
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=iq3_xxs
+^blk\.3\.attn_v\.weight$=q6_k
+^blk\.3\.ffn_down\.weight$=iq4_xs
+^blk\.3\.ffn_gate\.weight$=iq3_xxs
+^blk\.3\.ffn_up\.weight$=iq4_xs
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=iq3_xxs
+^blk\.4\.attn_v\.weight$=q6_k
+^blk\.4\.ffn_down\.weight$=iq4_xs
+^blk\.4\.ffn_gate\.weight$=iq3_xxs
+^blk\.4\.ffn_up\.weight$=iq4_xs
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=iq3_xxs
+^blk\.5\.attn_v\.weight$=q6_k
+^blk\.5\.ffn_down\.weight$=iq4_xs
+^blk\.5\.ffn_gate\.weight$=iq3_xxs
+^blk\.5\.ffn_up\.weight$=iq4_xs
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=iq3_xxs
+^blk\.6\.attn_v\.weight$=q6_k
+^blk\.6\.ffn_down\.weight$=iq4_xs
+^blk\.6\.ffn_gate\.weight$=iq3_xxs
+^blk\.6\.ffn_up\.weight$=iq4_xs
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=iq3_xxs
+^blk\.7\.attn_v\.weight$=q6_k
+^blk\.7\.ffn_down\.weight$=iq4_xs
+^blk\.7\.ffn_gate\.weight$=iq3_xxs
+^blk\.7\.ffn_up\.weight$=iq4_xs
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=iq3_xxs
+^blk\.8\.attn_v\.weight$=q6_k
+^blk\.8\.ffn_down\.weight$=iq4_xs
+^blk\.8\.ffn_gate\.weight$=iq3_xxs
+^blk\.8\.ffn_up\.weight$=iq4_xs
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=iq3_xxs
+^blk\.9\.attn_v\.weight$=q6_k
+^blk\.9\.ffn_down\.weight$=iq4_xs
+^blk\.9\.ffn_gate\.weight$=iq3_xxs
+^blk\.9\.ffn_up\.weight$=iq4_xs
+^blk\.10\.attn_k\.weight$=q5_k
+^blk\.10\.attn_output\.weight$=iq4_xs
+^blk\.10\.attn_q\.weight$=iq3_xxs
+^blk\.10\.attn_v\.weight$=q6_k
+^blk\.10\.ffn_down\.weight$=iq3_xxs
+^blk\.10\.ffn_gate\.weight$=iq3_xxs
+^blk\.10\.ffn_up\.weight$=iq3_xxs
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=iq4_xs
+^blk\.11\.attn_q\.weight$=iq3_xxs
+^blk\.11\.attn_v\.weight$=q6_k
+^blk\.11\.ffn_down\.weight$=iq3_xxs
+^blk\.11\.ffn_gate\.weight$=iq3_xxs
+^blk\.11\.ffn_up\.weight$=iq3_xxs
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=iq4_xs
+^blk\.12\.attn_q\.weight$=iq3_xxs
+^blk\.12\.attn_v\.weight$=q6_k
+^blk\.12\.ffn_down\.weight$=iq3_xxs
+^blk\.12\.ffn_gate\.weight$=iq3_xxs
+^blk\.12\.ffn_up\.weight$=iq3_xxs
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=iq4_xs
+^blk\.13\.attn_q\.weight$=iq3_xxs
+^blk\.13\.attn_v\.weight$=q6_k
+^blk\.13\.ffn_down\.weight$=iq3_xxs
+^blk\.13\.ffn_gate\.weight$=iq3_xxs
+^blk\.13\.ffn_up\.weight$=iq3_xxs
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=iq4_xs
+^blk\.14\.attn_q\.weight$=iq3_xxs
+^blk\.14\.attn_v\.weight$=q5_k
+^blk\.14\.ffn_down\.weight$=iq3_xxs
+^blk\.14\.ffn_gate\.weight$=iq3_xxs
+^blk\.14\.ffn_up\.weight$=iq3_xxs
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=iq4_xs
+^blk\.15\.attn_q\.weight$=iq3_xxs
+^blk\.15\.attn_v\.weight$=q5_k
+^blk\.15\.ffn_down\.weight$=iq3_xxs
+^blk\.15\.ffn_gate\.weight$=iq3_xxs
+^blk\.15\.ffn_up\.weight$=iq3_xxs
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=iq4_xs
+^blk\.16\.attn_q\.weight$=iq3_xxs
+^blk\.16\.attn_v\.weight$=q5_k
+^blk\.16\.ffn_down\.weight$=iq3_xxs
+^blk\.16\.ffn_gate\.weight$=iq3_xxs
+^blk\.16\.ffn_up\.weight$=iq3_xxs
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=iq4_xs
+^blk\.17\.attn_q\.weight$=iq3_xxs
+^blk\.17\.attn_v\.weight$=q5_k
+^blk\.17\.ffn_down\.weight$=iq3_xxs
+^blk\.17\.ffn_gate\.weight$=iq3_xxs
+^blk\.17\.ffn_up\.weight$=iq3_xxs
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=iq4_xs
+^blk\.18\.attn_q\.weight$=iq3_xxs
+^blk\.18\.attn_v\.weight$=q5_k
+^blk\.18\.ffn_down\.weight$=iq3_xxs
+^blk\.18\.ffn_gate\.weight$=iq3_xxs
+^blk\.18\.ffn_up\.weight$=iq3_xxs
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=iq4_xs
+^blk\.19\.attn_q\.weight$=iq3_xxs
+^blk\.19\.attn_v\.weight$=q5_k
+^blk\.19\.ffn_down\.weight$=iq3_xxs
+^blk\.19\.ffn_gate\.weight$=iq3_xxs
+^blk\.19\.ffn_up\.weight$=iq3_xxs
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=iq4_xs
+^blk\.20\.attn_q\.weight$=iq3_xxs
+^blk\.20\.attn_v\.weight$=q5_k
+^blk\.20\.ffn_down\.weight$=iq3_xxs
+^blk\.20\.ffn_gate\.weight$=iq3_xxs
+^blk\.20\.ffn_up\.weight$=iq3_xxs
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=iq4_xs
+^blk\.21\.attn_q\.weight$=iq3_xxs
+^blk\.21\.attn_v\.weight$=q5_k
+^blk\.21\.ffn_down\.weight$=iq3_xxs
+^blk\.21\.ffn_gate\.weight$=iq3_xxs
+^blk\.21\.ffn_up\.weight$=iq3_xxs
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=iq4_xs
+^blk\.22\.attn_q\.weight$=iq3_xxs
+^blk\.22\.attn_v\.weight$=q5_k
+^blk\.22\.ffn_down\.weight$=iq3_xxs
+^blk\.22\.ffn_gate\.weight$=iq3_xxs
+^blk\.22\.ffn_up\.weight$=iq3_xxs
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=iq4_xs
+^blk\.23\.attn_q\.weight$=iq3_xxs
+^blk\.23\.attn_v\.weight$=q5_k
+^blk\.23\.ffn_down\.weight$=iq3_xxs
+^blk\.23\.ffn_gate\.weight$=iq3_xxs
+^blk\.23\.ffn_up\.weight$=iq3_xxs
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=iq4_xs
+^blk\.24\.attn_q\.weight$=iq3_xxs
+^blk\.24\.attn_v\.weight$=q5_k
+^blk\.24\.ffn_down\.weight$=iq3_xxs
+^blk\.24\.ffn_gate\.weight$=iq3_xxs
+^blk\.24\.ffn_up\.weight$=iq3_xxs
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=iq4_xs
+^blk\.25\.attn_q\.weight$=iq3_xxs
+^blk\.25\.attn_v\.weight$=q5_k
+^blk\.25\.ffn_down\.weight$=iq3_xxs
+^blk\.25\.ffn_gate\.weight$=iq3_xxs
+^blk\.25\.ffn_up\.weight$=iq3_xxs
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=iq4_xs
+^blk\.26\.attn_q\.weight$=iq3_xxs
+^blk\.26\.attn_v\.weight$=q5_k
+^blk\.26\.ffn_down\.weight$=iq3_xxs
+^blk\.26\.ffn_gate\.weight$=iq3_xxs
+^blk\.26\.ffn_up\.weight$=iq3_xxs
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=iq4_xs
+^blk\.27\.attn_q\.weight$=iq3_xxs
+^blk\.27\.attn_v\.weight$=q5_k
+^blk\.27\.ffn_down\.weight$=iq3_xxs
+^blk\.27\.ffn_gate\.weight$=iq3_xxs
+^blk\.27\.ffn_up\.weight$=iq3_xxs
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=iq4_xs
+^blk\.28\.attn_q\.weight$=iq3_xxs
+^blk\.28\.attn_v\.weight$=q5_k
+^blk\.28\.ffn_down\.weight$=iq3_xxs
+^blk\.28\.ffn_gate\.weight$=iq3_xxs
+^blk\.28\.ffn_up\.weight$=iq3_xxs
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=iq4_xs
+^blk\.29\.attn_q\.weight$=iq3_xxs
+^blk\.29\.attn_v\.weight$=q5_k
+^blk\.29\.ffn_down\.weight$=iq3_xxs
+^blk\.29\.ffn_gate\.weight$=iq3_xxs
+^blk\.29\.ffn_up\.weight$=iq3_xxs
+^blk\.30\.attn_k\.weight$=q4_k
+^blk\.30\.attn_output\.weight$=iq3_xxs
+^blk\.30\.attn_q\.weight$=iq3_xxs
+^blk\.30\.attn_v\.weight$=q5_k
+^blk\.30\.ffn_down\.weight$=iq3_xxs
+^blk\.30\.ffn_gate\.weight$=iq3_xxs
+^blk\.30\.ffn_up\.weight$=iq3_xxs
+^blk\.31\.attn_k\.weight$=q4_k
+^blk\.31\.attn_output\.weight$=iq3_xxs
+^blk\.31\.attn_q\.weight$=iq3_xxs
+^blk\.31\.attn_v\.weight$=q5_k
+^blk\.31\.ffn_down\.weight$=iq3_xxs
+^blk\.31\.ffn_gate\.weight$=iq3_xxs
+^blk\.31\.ffn_up\.weight$=iq3_xxs
+^blk\.32\.attn_k\.weight$=q4_k
+^blk\.32\.attn_output\.weight$=iq3_xxs
+^blk\.32\.attn_q\.weight$=iq3_xxs
+^blk\.32\.attn_v\.weight$=q5_k
+^blk\.32\.ffn_down\.weight$=iq3_xxs
+^blk\.32\.ffn_gate\.weight$=iq3_xxs
+^blk\.32\.ffn_up\.weight$=iq3_xxs
+^blk\.33\.attn_k\.weight$=q4_k
+^blk\.33\.attn_output\.weight$=iq3_xxs
+^blk\.33\.attn_q\.weight$=iq3_xxs
+^blk\.33\.attn_v\.weight$=q5_k
+^blk\.33\.ffn_down\.weight$=iq3_xxs
+^blk\.33\.ffn_gate\.weight$=iq3_xxs
+^blk\.33\.ffn_up\.weight$=iq3_xxs
+^blk\.34\.attn_k\.weight$=q4_k
+^blk\.34\.attn_output\.weight$=iq3_xxs
+^blk\.34\.attn_q\.weight$=iq3_xxs
+^blk\.34\.attn_v\.weight$=q5_k
+^blk\.34\.ffn_down\.weight$=iq3_xxs
+^blk\.34\.ffn_gate\.weight$=iq3_xxs
+^blk\.34\.ffn_up\.weight$=iq3_xxs
+^blk\.35\.attn_k\.weight$=q4_k
+^blk\.35\.attn_output\.weight$=iq3_xxs
+^blk\.35\.attn_q\.weight$=iq3_xxs
+^blk\.35\.attn_v\.weight$=q5_k
+^blk\.35\.ffn_down\.weight$=iq3_xxs
+^blk\.35\.ffn_gate\.weight$=iq3_xxs
+^blk\.35\.ffn_up\.weight$=iq3_xxs
+^blk\.36\.attn_k\.weight$=q4_k
+^blk\.36\.attn_output\.weight$=iq3_xxs
+^blk\.36\.attn_q\.weight$=iq3_xxs
+^blk\.36\.attn_v\.weight$=q5_k
+^blk\.36\.ffn_down\.weight$=iq3_xxs
+^blk\.36\.ffn_gate\.weight$=iq3_xxs
+^blk\.36\.ffn_up\.weight$=iq3_xxs
+^blk\.37\.attn_k\.weight$=q4_k
+^blk\.37\.attn_output\.weight$=iq3_xxs
+^blk\.37\.attn_q\.weight$=iq3_xxs
+^blk\.37\.attn_v\.weight$=q5_k
+^blk\.37\.ffn_down\.weight$=iq3_xxs
+^blk\.37\.ffn_gate\.weight$=iq3_xxs
+^blk\.37\.ffn_up\.weight$=iq3_xxs
+^blk\.38\.attn_k\.weight$=q4_k
+^blk\.38\.attn_output\.weight$=iq3_xxs
+^blk\.38\.attn_q\.weight$=iq3_xxs
+^blk\.38\.attn_v\.weight$=q5_k
+^blk\.38\.ffn_down\.weight$=iq3_xxs
+^blk\.38\.ffn_gate\.weight$=iq3_xxs
+^blk\.38\.ffn_up\.weight$=iq3_xxs
+^blk\.39\.attn_k\.weight$=q4_k
+^blk\.39\.attn_output\.weight$=iq3_xxs
+^blk\.39\.attn_q\.weight$=iq3_xxs
+^blk\.39\.attn_v\.weight$=q5_k
+^blk\.39\.ffn_down\.weight$=iq4_xs
+^blk\.39\.ffn_gate\.weight$=iq3_xxs
+^blk\.39\.ffn_up\.weight$=q4_k

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ4_NL.layout.json

@@ -0,0 +1,87 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ4_NL",
+  "base_type": "iq4_nl",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0732,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0502,
+      "type": "iq4_nl"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "iq4_nl"
+  },
+  "bytes": {
+    "heuristic_body+embd": 8694848280,
+    "map_body+embd": 9267510040,
+    "ratio": 1.0659,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 4.961,
+  "body_bpw_heuristic": 4.615,
+  "total_params": 14765593600,
+  "file_bytes_pred": 9269209880,
+  "file_bpw_pred": 5.022,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 23,
+      "q6_k": 17
+    },
+    "attn_output": {
+      "q6_k": 28,
+      "iq4_nl": 9,
+      "q5_k": 2,
+      "q8_0": 1
+    },
+    "attn_q": {
+      "iq4_nl": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "iq4_nl": 30,
+      "q5_k": 8,
+      "q6_k": 2
+    },
+    "ffn_gate": {
+      "iq4_nl": 40
+    },
+    "ffn_up": {
+      "iq4_nl": 30,
+      "q6_k": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ4_NL",
+  "quantize_ftype": "IQ4_NL",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ4_NL.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=iq4_nl
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q5_k
+^blk\.0\.attn_q\.weight$=iq4_nl
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=iq4_nl
+^blk\.0\.ffn_gate\.weight$=iq4_nl
+^blk\.0\.ffn_up\.weight$=iq4_nl
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=iq4_nl
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q6_k
+^blk\.1\.ffn_gate\.weight$=iq4_nl
+^blk\.1\.ffn_up\.weight$=q6_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q6_k
+^blk\.2\.attn_q\.weight$=iq4_nl
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q5_k
+^blk\.2\.ffn_gate\.weight$=iq4_nl
+^blk\.2\.ffn_up\.weight$=q6_k
+^blk\.3\.attn_k\.weight$=q6_k
+^blk\.3\.attn_output\.weight$=q6_k
+^blk\.3\.attn_q\.weight$=iq4_nl
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q5_k
+^blk\.3\.ffn_gate\.weight$=iq4_nl
+^blk\.3\.ffn_up\.weight$=q6_k
+^blk\.4\.attn_k\.weight$=q6_k
+^blk\.4\.attn_output\.weight$=q6_k
+^blk\.4\.attn_q\.weight$=iq4_nl
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q5_k
+^blk\.4\.ffn_gate\.weight$=iq4_nl
+^blk\.4\.ffn_up\.weight$=q6_k
+^blk\.5\.attn_k\.weight$=q6_k
+^blk\.5\.attn_output\.weight$=q6_k
+^blk\.5\.attn_q\.weight$=iq4_nl
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q5_k
+^blk\.5\.ffn_gate\.weight$=iq4_nl
+^blk\.5\.ffn_up\.weight$=q6_k
+^blk\.6\.attn_k\.weight$=q6_k
+^blk\.6\.attn_output\.weight$=q6_k
+^blk\.6\.attn_q\.weight$=iq4_nl
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q5_k
+^blk\.6\.ffn_gate\.weight$=iq4_nl
+^blk\.6\.ffn_up\.weight$=q6_k
+^blk\.7\.attn_k\.weight$=q6_k
+^blk\.7\.attn_output\.weight$=q6_k
+^blk\.7\.attn_q\.weight$=iq4_nl
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q5_k
+^blk\.7\.ffn_gate\.weight$=iq4_nl
+^blk\.7\.ffn_up\.weight$=q6_k
+^blk\.8\.attn_k\.weight$=q6_k
+^blk\.8\.attn_output\.weight$=q6_k
+^blk\.8\.attn_q\.weight$=iq4_nl
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q5_k
+^blk\.8\.ffn_gate\.weight$=iq4_nl
+^blk\.8\.ffn_up\.weight$=q6_k
+^blk\.9\.attn_k\.weight$=q6_k
+^blk\.9\.attn_output\.weight$=q6_k
+^blk\.9\.attn_q\.weight$=iq4_nl
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q5_k
+^blk\.9\.ffn_gate\.weight$=iq4_nl
+^blk\.9\.ffn_up\.weight$=q6_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q6_k
+^blk\.10\.attn_q\.weight$=iq4_nl
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=iq4_nl
+^blk\.10\.ffn_gate\.weight$=iq4_nl
+^blk\.10\.ffn_up\.weight$=iq4_nl
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q6_k
+^blk\.11\.attn_q\.weight$=iq4_nl
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=iq4_nl
+^blk\.11\.ffn_gate\.weight$=iq4_nl
+^blk\.11\.ffn_up\.weight$=iq4_nl
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q6_k
+^blk\.12\.attn_q\.weight$=iq4_nl
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=iq4_nl
+^blk\.12\.ffn_gate\.weight$=iq4_nl
+^blk\.12\.ffn_up\.weight$=iq4_nl
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q6_k
+^blk\.13\.attn_q\.weight$=iq4_nl
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=iq4_nl
+^blk\.13\.ffn_gate\.weight$=iq4_nl
+^blk\.13\.ffn_up\.weight$=iq4_nl
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q6_k
+^blk\.14\.attn_q\.weight$=iq4_nl
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=iq4_nl
+^blk\.14\.ffn_gate\.weight$=iq4_nl
+^blk\.14\.ffn_up\.weight$=iq4_nl
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q6_k
+^blk\.15\.attn_q\.weight$=iq4_nl
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=iq4_nl
+^blk\.15\.ffn_gate\.weight$=iq4_nl
+^blk\.15\.ffn_up\.weight$=iq4_nl
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q6_k
+^blk\.16\.attn_q\.weight$=iq4_nl
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=iq4_nl
+^blk\.16\.ffn_gate\.weight$=iq4_nl
+^blk\.16\.ffn_up\.weight$=iq4_nl
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q6_k
+^blk\.17\.attn_q\.weight$=iq4_nl
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=iq4_nl
+^blk\.17\.ffn_gate\.weight$=iq4_nl
+^blk\.17\.ffn_up\.weight$=iq4_nl
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q6_k
+^blk\.18\.attn_q\.weight$=iq4_nl
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=iq4_nl
+^blk\.18\.ffn_gate\.weight$=iq4_nl
+^blk\.18\.ffn_up\.weight$=iq4_nl
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q6_k
+^blk\.19\.attn_q\.weight$=iq4_nl
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=iq4_nl
+^blk\.19\.ffn_gate\.weight$=iq4_nl
+^blk\.19\.ffn_up\.weight$=iq4_nl
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q6_k
+^blk\.20\.attn_q\.weight$=iq4_nl
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=iq4_nl
+^blk\.20\.ffn_gate\.weight$=iq4_nl
+^blk\.20\.ffn_up\.weight$=iq4_nl
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q6_k
+^blk\.21\.attn_q\.weight$=iq4_nl
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=iq4_nl
+^blk\.21\.ffn_gate\.weight$=iq4_nl
+^blk\.21\.ffn_up\.weight$=iq4_nl
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q6_k
+^blk\.22\.attn_q\.weight$=iq4_nl
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=iq4_nl
+^blk\.22\.ffn_gate\.weight$=iq4_nl
+^blk\.22\.ffn_up\.weight$=iq4_nl
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q6_k
+^blk\.23\.attn_q\.weight$=iq4_nl
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=iq4_nl
+^blk\.23\.ffn_gate\.weight$=iq4_nl
+^blk\.23\.ffn_up\.weight$=iq4_nl
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q6_k
+^blk\.24\.attn_q\.weight$=iq4_nl
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=iq4_nl
+^blk\.24\.ffn_gate\.weight$=iq4_nl
+^blk\.24\.ffn_up\.weight$=iq4_nl
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q6_k
+^blk\.25\.attn_q\.weight$=iq4_nl
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=iq4_nl
+^blk\.25\.ffn_gate\.weight$=iq4_nl
+^blk\.25\.ffn_up\.weight$=iq4_nl
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q6_k
+^blk\.26\.attn_q\.weight$=iq4_nl
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=iq4_nl
+^blk\.26\.ffn_gate\.weight$=iq4_nl
+^blk\.26\.ffn_up\.weight$=iq4_nl
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q6_k
+^blk\.27\.attn_q\.weight$=iq4_nl
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=iq4_nl
+^blk\.27\.ffn_gate\.weight$=iq4_nl
+^blk\.27\.ffn_up\.weight$=iq4_nl
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q6_k
+^blk\.28\.attn_q\.weight$=iq4_nl
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=iq4_nl
+^blk\.28\.ffn_gate\.weight$=iq4_nl
+^blk\.28\.ffn_up\.weight$=iq4_nl
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q6_k
+^blk\.29\.attn_q\.weight$=iq4_nl
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=iq4_nl
+^blk\.29\.ffn_gate\.weight$=iq4_nl
+^blk\.29\.ffn_up\.weight$=iq4_nl
+^blk\.30\.attn_k\.weight$=q6_k
+^blk\.30\.attn_output\.weight$=iq4_nl
+^blk\.30\.attn_q\.weight$=iq4_nl
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=iq4_nl
+^blk\.30\.ffn_gate\.weight$=iq4_nl
+^blk\.30\.ffn_up\.weight$=iq4_nl
+^blk\.31\.attn_k\.weight$=q6_k
+^blk\.31\.attn_output\.weight$=iq4_nl
+^blk\.31\.attn_q\.weight$=iq4_nl
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=iq4_nl
+^blk\.31\.ffn_gate\.weight$=iq4_nl
+^blk\.31\.ffn_up\.weight$=iq4_nl
+^blk\.32\.attn_k\.weight$=q6_k
+^blk\.32\.attn_output\.weight$=iq4_nl
+^blk\.32\.attn_q\.weight$=iq4_nl
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=iq4_nl
+^blk\.32\.ffn_gate\.weight$=iq4_nl
+^blk\.32\.ffn_up\.weight$=iq4_nl
+^blk\.33\.attn_k\.weight$=q6_k
+^blk\.33\.attn_output\.weight$=iq4_nl
+^blk\.33\.attn_q\.weight$=iq4_nl
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=iq4_nl
+^blk\.33\.ffn_gate\.weight$=iq4_nl
+^blk\.33\.ffn_up\.weight$=iq4_nl
+^blk\.34\.attn_k\.weight$=q6_k
+^blk\.34\.attn_output\.weight$=iq4_nl
+^blk\.34\.attn_q\.weight$=iq4_nl
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=iq4_nl
+^blk\.34\.ffn_gate\.weight$=iq4_nl
+^blk\.34\.ffn_up\.weight$=iq4_nl
+^blk\.35\.attn_k\.weight$=q6_k
+^blk\.35\.attn_output\.weight$=iq4_nl
+^blk\.35\.attn_q\.weight$=iq4_nl
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=iq4_nl
+^blk\.35\.ffn_gate\.weight$=iq4_nl
+^blk\.35\.ffn_up\.weight$=iq4_nl
+^blk\.36\.attn_k\.weight$=q6_k
+^blk\.36\.attn_output\.weight$=iq4_nl
+^blk\.36\.attn_q\.weight$=iq4_nl
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=iq4_nl
+^blk\.36\.ffn_gate\.weight$=iq4_nl
+^blk\.36\.ffn_up\.weight$=iq4_nl
+^blk\.37\.attn_k\.weight$=q6_k
+^blk\.37\.attn_output\.weight$=iq4_nl
+^blk\.37\.attn_q\.weight$=iq4_nl
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=iq4_nl
+^blk\.37\.ffn_gate\.weight$=iq4_nl
+^blk\.37\.ffn_up\.weight$=iq4_nl
+^blk\.38\.attn_k\.weight$=q6_k
+^blk\.38\.attn_output\.weight$=iq4_nl
+^blk\.38\.attn_q\.weight$=iq4_nl
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=iq4_nl
+^blk\.38\.ffn_gate\.weight$=iq4_nl
+^blk\.38\.ffn_up\.weight$=iq4_nl
+^blk\.39\.attn_k\.weight$=q6_k
+^blk\.39\.attn_output\.weight$=q5_k
+^blk\.39\.attn_q\.weight$=iq4_nl
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q6_k
+^blk\.39\.ffn_gate\.weight$=iq4_nl
+^blk\.39\.ffn_up\.weight$=q6_k

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ4_XS.layout.json

@@ -0,0 +1,85 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "IQ4_XS",
+  "base_type": "iq4_xs",
+  "family": "iq",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0769,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0498,
+      "type": "iq4_xs"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "iq4_xs"
+  },
+  "bytes": {
+    "heuristic_body+embd": 8280641720,
+    "map_body+embd": 8346833080,
+    "ratio": 1.008,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.9,
+  "body_bpw": 4.419,
+  "body_bpw_heuristic": 4.378,
+  "total_params": 14765593600,
+  "file_bytes_pred": 8348532920,
+  "file_bpw_pred": 4.523,
+  "histogram": {
+    "attn_k": {
+      "q6_k": 38,
+      "q8_0": 2
+    },
+    "attn_output": {
+      "iq4_xs": 27,
+      "q6_k": 9,
+      "q5_k": 2,
+      "q4_k": 2
+    },
+    "attn_q": {
+      "iq4_xs": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "iq4_xs": 40
+    },
+    "ffn_gate": {
+      "iq4_xs": 40
+    },
+    "ffn_up": {
+      "iq4_xs": 39,
+      "q6_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "IQ4_XS",
+  "quantize_ftype": "IQ4_XS",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-IQ4_XS.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=iq4_xs
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=iq4_xs
+^blk\.0\.attn_q\.weight$=iq4_xs
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=iq4_xs
+^blk\.0\.ffn_gate\.weight$=iq4_xs
+^blk\.0\.ffn_up\.weight$=iq4_xs
+^blk\.1\.attn_k\.weight$=q6_k
+^blk\.1\.attn_output\.weight$=q6_k
+^blk\.1\.attn_q\.weight$=iq4_xs
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=iq4_xs
+^blk\.1\.ffn_gate\.weight$=iq4_xs
+^blk\.1\.ffn_up\.weight$=iq4_xs
+^blk\.2\.attn_k\.weight$=q6_k
+^blk\.2\.attn_output\.weight$=q6_k
+^blk\.2\.attn_q\.weight$=iq4_xs
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=iq4_xs
+^blk\.2\.ffn_gate\.weight$=iq4_xs
+^blk\.2\.ffn_up\.weight$=iq4_xs
+^blk\.3\.attn_k\.weight$=q6_k
+^blk\.3\.attn_output\.weight$=q6_k
+^blk\.3\.attn_q\.weight$=iq4_xs
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=iq4_xs
+^blk\.3\.ffn_gate\.weight$=iq4_xs
+^blk\.3\.ffn_up\.weight$=iq4_xs
+^blk\.4\.attn_k\.weight$=q6_k
+^blk\.4\.attn_output\.weight$=q6_k
+^blk\.4\.attn_q\.weight$=iq4_xs
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=iq4_xs
+^blk\.4\.ffn_gate\.weight$=iq4_xs
+^blk\.4\.ffn_up\.weight$=iq4_xs
+^blk\.5\.attn_k\.weight$=q6_k
+^blk\.5\.attn_output\.weight$=q6_k
+^blk\.5\.attn_q\.weight$=iq4_xs
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=iq4_xs
+^blk\.5\.ffn_gate\.weight$=iq4_xs
+^blk\.5\.ffn_up\.weight$=iq4_xs
+^blk\.6\.attn_k\.weight$=q6_k
+^blk\.6\.attn_output\.weight$=q6_k
+^blk\.6\.attn_q\.weight$=iq4_xs
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=iq4_xs
+^blk\.6\.ffn_gate\.weight$=iq4_xs
+^blk\.6\.ffn_up\.weight$=iq4_xs
+^blk\.7\.attn_k\.weight$=q6_k
+^blk\.7\.attn_output\.weight$=q6_k
+^blk\.7\.attn_q\.weight$=iq4_xs
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=iq4_xs
+^blk\.7\.ffn_gate\.weight$=iq4_xs
+^blk\.7\.ffn_up\.weight$=iq4_xs
+^blk\.8\.attn_k\.weight$=q6_k
+^blk\.8\.attn_output\.weight$=q6_k
+^blk\.8\.attn_q\.weight$=iq4_xs
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=iq4_xs
+^blk\.8\.ffn_gate\.weight$=iq4_xs
+^blk\.8\.ffn_up\.weight$=iq4_xs
+^blk\.9\.attn_k\.weight$=q6_k
+^blk\.9\.attn_output\.weight$=q6_k
+^blk\.9\.attn_q\.weight$=iq4_xs
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=iq4_xs
+^blk\.9\.ffn_gate\.weight$=iq4_xs
+^blk\.9\.ffn_up\.weight$=iq4_xs
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q5_k
+^blk\.10\.attn_q\.weight$=iq4_xs
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=iq4_xs
+^blk\.10\.ffn_gate\.weight$=iq4_xs
+^blk\.10\.ffn_up\.weight$=iq4_xs
+^blk\.11\.attn_k\.weight$=q6_k
+^blk\.11\.attn_output\.weight$=q5_k
+^blk\.11\.attn_q\.weight$=iq4_xs
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=iq4_xs
+^blk\.11\.ffn_gate\.weight$=iq4_xs
+^blk\.11\.ffn_up\.weight$=iq4_xs
+^blk\.12\.attn_k\.weight$=q6_k
+^blk\.12\.attn_output\.weight$=q4_k
+^blk\.12\.attn_q\.weight$=iq4_xs
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=iq4_xs
+^blk\.12\.ffn_gate\.weight$=iq4_xs
+^blk\.12\.ffn_up\.weight$=iq4_xs
+^blk\.13\.attn_k\.weight$=q6_k
+^blk\.13\.attn_output\.weight$=q4_k
+^blk\.13\.attn_q\.weight$=iq4_xs
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=iq4_xs
+^blk\.13\.ffn_gate\.weight$=iq4_xs
+^blk\.13\.ffn_up\.weight$=iq4_xs
+^blk\.14\.attn_k\.weight$=q6_k
+^blk\.14\.attn_output\.weight$=iq4_xs
+^blk\.14\.attn_q\.weight$=iq4_xs
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=iq4_xs
+^blk\.14\.ffn_gate\.weight$=iq4_xs
+^blk\.14\.ffn_up\.weight$=iq4_xs
+^blk\.15\.attn_k\.weight$=q6_k
+^blk\.15\.attn_output\.weight$=iq4_xs
+^blk\.15\.attn_q\.weight$=iq4_xs
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=iq4_xs
+^blk\.15\.ffn_gate\.weight$=iq4_xs
+^blk\.15\.ffn_up\.weight$=iq4_xs
+^blk\.16\.attn_k\.weight$=q6_k
+^blk\.16\.attn_output\.weight$=iq4_xs
+^blk\.16\.attn_q\.weight$=iq4_xs
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=iq4_xs
+^blk\.16\.ffn_gate\.weight$=iq4_xs
+^blk\.16\.ffn_up\.weight$=iq4_xs
+^blk\.17\.attn_k\.weight$=q6_k
+^blk\.17\.attn_output\.weight$=iq4_xs
+^blk\.17\.attn_q\.weight$=iq4_xs
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=iq4_xs
+^blk\.17\.ffn_gate\.weight$=iq4_xs
+^blk\.17\.ffn_up\.weight$=iq4_xs
+^blk\.18\.attn_k\.weight$=q6_k
+^blk\.18\.attn_output\.weight$=iq4_xs
+^blk\.18\.attn_q\.weight$=iq4_xs
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=iq4_xs
+^blk\.18\.ffn_gate\.weight$=iq4_xs
+^blk\.18\.ffn_up\.weight$=iq4_xs
+^blk\.19\.attn_k\.weight$=q6_k
+^blk\.19\.attn_output\.weight$=iq4_xs
+^blk\.19\.attn_q\.weight$=iq4_xs
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=iq4_xs
+^blk\.19\.ffn_gate\.weight$=iq4_xs
+^blk\.19\.ffn_up\.weight$=iq4_xs
+^blk\.20\.attn_k\.weight$=q6_k
+^blk\.20\.attn_output\.weight$=iq4_xs
+^blk\.20\.attn_q\.weight$=iq4_xs
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=iq4_xs
+^blk\.20\.ffn_gate\.weight$=iq4_xs
+^blk\.20\.ffn_up\.weight$=iq4_xs
+^blk\.21\.attn_k\.weight$=q6_k
+^blk\.21\.attn_output\.weight$=iq4_xs
+^blk\.21\.attn_q\.weight$=iq4_xs
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=iq4_xs
+^blk\.21\.ffn_gate\.weight$=iq4_xs
+^blk\.21\.ffn_up\.weight$=iq4_xs
+^blk\.22\.attn_k\.weight$=q6_k
+^blk\.22\.attn_output\.weight$=iq4_xs
+^blk\.22\.attn_q\.weight$=iq4_xs
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=iq4_xs
+^blk\.22\.ffn_gate\.weight$=iq4_xs
+^blk\.22\.ffn_up\.weight$=iq4_xs
+^blk\.23\.attn_k\.weight$=q6_k
+^blk\.23\.attn_output\.weight$=iq4_xs
+^blk\.23\.attn_q\.weight$=iq4_xs
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=iq4_xs
+^blk\.23\.ffn_gate\.weight$=iq4_xs
+^blk\.23\.ffn_up\.weight$=iq4_xs
+^blk\.24\.attn_k\.weight$=q6_k
+^blk\.24\.attn_output\.weight$=iq4_xs
+^blk\.24\.attn_q\.weight$=iq4_xs
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=iq4_xs
+^blk\.24\.ffn_gate\.weight$=iq4_xs
+^blk\.24\.ffn_up\.weight$=iq4_xs
+^blk\.25\.attn_k\.weight$=q6_k
+^blk\.25\.attn_output\.weight$=iq4_xs
+^blk\.25\.attn_q\.weight$=iq4_xs
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=iq4_xs
+^blk\.25\.ffn_gate\.weight$=iq4_xs
+^blk\.25\.ffn_up\.weight$=iq4_xs
+^blk\.26\.attn_k\.weight$=q6_k
+^blk\.26\.attn_output\.weight$=iq4_xs
+^blk\.26\.attn_q\.weight$=iq4_xs
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=iq4_xs
+^blk\.26\.ffn_gate\.weight$=iq4_xs
+^blk\.26\.ffn_up\.weight$=iq4_xs
+^blk\.27\.attn_k\.weight$=q6_k
+^blk\.27\.attn_output\.weight$=iq4_xs
+^blk\.27\.attn_q\.weight$=iq4_xs
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=iq4_xs
+^blk\.27\.ffn_gate\.weight$=iq4_xs
+^blk\.27\.ffn_up\.weight$=iq4_xs
+^blk\.28\.attn_k\.weight$=q6_k
+^blk\.28\.attn_output\.weight$=iq4_xs
+^blk\.28\.attn_q\.weight$=iq4_xs
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=iq4_xs
+^blk\.28\.ffn_gate\.weight$=iq4_xs
+^blk\.28\.ffn_up\.weight$=iq4_xs
+^blk\.29\.attn_k\.weight$=q6_k
+^blk\.29\.attn_output\.weight$=iq4_xs
+^blk\.29\.attn_q\.weight$=iq4_xs
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=iq4_xs
+^blk\.29\.ffn_gate\.weight$=iq4_xs
+^blk\.29\.ffn_up\.weight$=iq4_xs
+^blk\.30\.attn_k\.weight$=q6_k
+^blk\.30\.attn_output\.weight$=iq4_xs
+^blk\.30\.attn_q\.weight$=iq4_xs
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=iq4_xs
+^blk\.30\.ffn_gate\.weight$=iq4_xs
+^blk\.30\.ffn_up\.weight$=iq4_xs
+^blk\.31\.attn_k\.weight$=q6_k
+^blk\.31\.attn_output\.weight$=iq4_xs
+^blk\.31\.attn_q\.weight$=iq4_xs
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=iq4_xs
+^blk\.31\.ffn_gate\.weight$=iq4_xs
+^blk\.31\.ffn_up\.weight$=iq4_xs
+^blk\.32\.attn_k\.weight$=q6_k
+^blk\.32\.attn_output\.weight$=iq4_xs
+^blk\.32\.attn_q\.weight$=iq4_xs
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=iq4_xs
+^blk\.32\.ffn_gate\.weight$=iq4_xs
+^blk\.32\.ffn_up\.weight$=iq4_xs
+^blk\.33\.attn_k\.weight$=q6_k
+^blk\.33\.attn_output\.weight$=iq4_xs
+^blk\.33\.attn_q\.weight$=iq4_xs
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=iq4_xs
+^blk\.33\.ffn_gate\.weight$=iq4_xs
+^blk\.33\.ffn_up\.weight$=iq4_xs
+^blk\.34\.attn_k\.weight$=q6_k
+^blk\.34\.attn_output\.weight$=iq4_xs
+^blk\.34\.attn_q\.weight$=iq4_xs
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=iq4_xs
+^blk\.34\.ffn_gate\.weight$=iq4_xs
+^blk\.34\.ffn_up\.weight$=iq4_xs
+^blk\.35\.attn_k\.weight$=q6_k
+^blk\.35\.attn_output\.weight$=iq4_xs
+^blk\.35\.attn_q\.weight$=iq4_xs
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=iq4_xs
+^blk\.35\.ffn_gate\.weight$=iq4_xs
+^blk\.35\.ffn_up\.weight$=iq4_xs
+^blk\.36\.attn_k\.weight$=q6_k
+^blk\.36\.attn_output\.weight$=iq4_xs
+^blk\.36\.attn_q\.weight$=iq4_xs
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=iq4_xs
+^blk\.36\.ffn_gate\.weight$=iq4_xs
+^blk\.36\.ffn_up\.weight$=iq4_xs
+^blk\.37\.attn_k\.weight$=q6_k
+^blk\.37\.attn_output\.weight$=iq4_xs
+^blk\.37\.attn_q\.weight$=iq4_xs
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=iq4_xs
+^blk\.37\.ffn_gate\.weight$=iq4_xs
+^blk\.37\.ffn_up\.weight$=iq4_xs
+^blk\.38\.attn_k\.weight$=q6_k
+^blk\.38\.attn_output\.weight$=iq4_xs
+^blk\.38\.attn_q\.weight$=iq4_xs
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=iq4_xs
+^blk\.38\.ffn_gate\.weight$=iq4_xs
+^blk\.38\.ffn_up\.weight$=iq4_xs
+^blk\.39\.attn_k\.weight$=q6_k
+^blk\.39\.attn_output\.weight$=iq4_xs
+^blk\.39\.attn_q\.weight$=iq4_xs
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=iq4_xs
+^blk\.39\.ffn_gate\.weight$=iq4_xs
+^blk\.39\.ffn_up\.weight$=q6_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q2_K.layout.json

@@ -0,0 +1,86 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q2_K",
+  "base_type": "q2_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.1061,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0424,
+      "type": "q3_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q3_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 6002814280,
+    "map_body+embd": 5785010920,
+    "ratio": 0.9637,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 2.978,
+  "body_bpw_heuristic": 3.095,
+  "total_params": 14765593600,
+  "file_bytes_pred": 5786710760,
+  "file_bpw_pred": 3.135,
+  "histogram": {
+    "attn_k": {
+      "q4_k": 40
+    },
+    "attn_output": {
+      "q4_k": 29,
+      "q2_k": 10,
+      "q3_k": 1
+    },
+    "attn_q": {
+      "q2_k": 40
+    },
+    "attn_v": {
+      "q5_k": 29,
+      "q4_k": 11
+    },
+    "ffn_down": {
+      "q2_k": 33,
+      "q3_k": 6,
+      "q4_k": 1
+    },
+    "ffn_gate": {
+      "q2_k": 40
+    },
+    "ffn_up": {
+      "q2_k": 30,
+      "q4_k": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q2_K",
+  "quantize_ftype": "Q2_K",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q2_K.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q3_k
+^blk\.0\.attn_k\.weight$=q4_k
+^blk\.0\.attn_output\.weight$=q3_k
+^blk\.0\.attn_q\.weight$=q2_k
+^blk\.0\.attn_v\.weight$=q4_k
+^blk\.0\.ffn_down\.weight$=q2_k
+^blk\.0\.ffn_gate\.weight$=q2_k
+^blk\.0\.ffn_up\.weight$=q2_k
+^blk\.1\.attn_k\.weight$=q4_k
+^blk\.1\.attn_output\.weight$=q4_k
+^blk\.1\.attn_q\.weight$=q2_k
+^blk\.1\.attn_v\.weight$=q5_k
+^blk\.1\.ffn_down\.weight$=q3_k
+^blk\.1\.ffn_gate\.weight$=q2_k
+^blk\.1\.ffn_up\.weight$=q4_k
+^blk\.2\.attn_k\.weight$=q4_k
+^blk\.2\.attn_output\.weight$=q4_k
+^blk\.2\.attn_q\.weight$=q2_k
+^blk\.2\.attn_v\.weight$=q5_k
+^blk\.2\.ffn_down\.weight$=q3_k
+^blk\.2\.ffn_gate\.weight$=q2_k
+^blk\.2\.ffn_up\.weight$=q4_k
+^blk\.3\.attn_k\.weight$=q4_k
+^blk\.3\.attn_output\.weight$=q4_k
+^blk\.3\.attn_q\.weight$=q2_k
+^blk\.3\.attn_v\.weight$=q5_k
+^blk\.3\.ffn_down\.weight$=q3_k
+^blk\.3\.ffn_gate\.weight$=q2_k
+^blk\.3\.ffn_up\.weight$=q4_k
+^blk\.4\.attn_k\.weight$=q4_k
+^blk\.4\.attn_output\.weight$=q4_k
+^blk\.4\.attn_q\.weight$=q2_k
+^blk\.4\.attn_v\.weight$=q5_k
+^blk\.4\.ffn_down\.weight$=q3_k
+^blk\.4\.ffn_gate\.weight$=q2_k
+^blk\.4\.ffn_up\.weight$=q4_k
+^blk\.5\.attn_k\.weight$=q4_k
+^blk\.5\.attn_output\.weight$=q4_k
+^blk\.5\.attn_q\.weight$=q2_k
+^blk\.5\.attn_v\.weight$=q5_k
+^blk\.5\.ffn_down\.weight$=q3_k
+^blk\.5\.ffn_gate\.weight$=q2_k
+^blk\.5\.ffn_up\.weight$=q4_k
+^blk\.6\.attn_k\.weight$=q4_k
+^blk\.6\.attn_output\.weight$=q4_k
+^blk\.6\.attn_q\.weight$=q2_k
+^blk\.6\.attn_v\.weight$=q5_k
+^blk\.6\.ffn_down\.weight$=q3_k
+^blk\.6\.ffn_gate\.weight$=q2_k
+^blk\.6\.ffn_up\.weight$=q4_k
+^blk\.7\.attn_k\.weight$=q4_k
+^blk\.7\.attn_output\.weight$=q4_k
+^blk\.7\.attn_q\.weight$=q2_k
+^blk\.7\.attn_v\.weight$=q5_k
+^blk\.7\.ffn_down\.weight$=q2_k
+^blk\.7\.ffn_gate\.weight$=q2_k
+^blk\.7\.ffn_up\.weight$=q4_k
+^blk\.8\.attn_k\.weight$=q4_k
+^blk\.8\.attn_output\.weight$=q4_k
+^blk\.8\.attn_q\.weight$=q2_k
+^blk\.8\.attn_v\.weight$=q5_k
+^blk\.8\.ffn_down\.weight$=q2_k
+^blk\.8\.ffn_gate\.weight$=q2_k
+^blk\.8\.ffn_up\.weight$=q4_k
+^blk\.9\.attn_k\.weight$=q4_k
+^blk\.9\.attn_output\.weight$=q4_k
+^blk\.9\.attn_q\.weight$=q2_k
+^blk\.9\.attn_v\.weight$=q5_k
+^blk\.9\.ffn_down\.weight$=q2_k
+^blk\.9\.ffn_gate\.weight$=q2_k
+^blk\.9\.ffn_up\.weight$=q4_k
+^blk\.10\.attn_k\.weight$=q4_k
+^blk\.10\.attn_output\.weight$=q4_k
+^blk\.10\.attn_q\.weight$=q2_k
+^blk\.10\.attn_v\.weight$=q5_k
+^blk\.10\.ffn_down\.weight$=q2_k
+^blk\.10\.ffn_gate\.weight$=q2_k
+^blk\.10\.ffn_up\.weight$=q2_k
+^blk\.11\.attn_k\.weight$=q4_k
+^blk\.11\.attn_output\.weight$=q4_k
+^blk\.11\.attn_q\.weight$=q2_k
+^blk\.11\.attn_v\.weight$=q5_k
+^blk\.11\.ffn_down\.weight$=q2_k
+^blk\.11\.ffn_gate\.weight$=q2_k
+^blk\.11\.ffn_up\.weight$=q2_k
+^blk\.12\.attn_k\.weight$=q4_k
+^blk\.12\.attn_output\.weight$=q4_k
+^blk\.12\.attn_q\.weight$=q2_k
+^blk\.12\.attn_v\.weight$=q5_k
+^blk\.12\.ffn_down\.weight$=q2_k
+^blk\.12\.ffn_gate\.weight$=q2_k
+^blk\.12\.ffn_up\.weight$=q2_k
+^blk\.13\.attn_k\.weight$=q4_k
+^blk\.13\.attn_output\.weight$=q4_k
+^blk\.13\.attn_q\.weight$=q2_k
+^blk\.13\.attn_v\.weight$=q5_k
+^blk\.13\.ffn_down\.weight$=q2_k
+^blk\.13\.ffn_gate\.weight$=q2_k
+^blk\.13\.ffn_up\.weight$=q2_k
+^blk\.14\.attn_k\.weight$=q4_k
+^blk\.14\.attn_output\.weight$=q4_k
+^blk\.14\.attn_q\.weight$=q2_k
+^blk\.14\.attn_v\.weight$=q5_k
+^blk\.14\.ffn_down\.weight$=q2_k
+^blk\.14\.ffn_gate\.weight$=q2_k
+^blk\.14\.ffn_up\.weight$=q2_k
+^blk\.15\.attn_k\.weight$=q4_k
+^blk\.15\.attn_output\.weight$=q4_k
+^blk\.15\.attn_q\.weight$=q2_k
+^blk\.15\.attn_v\.weight$=q5_k
+^blk\.15\.ffn_down\.weight$=q2_k
+^blk\.15\.ffn_gate\.weight$=q2_k
+^blk\.15\.ffn_up\.weight$=q2_k
+^blk\.16\.attn_k\.weight$=q4_k
+^blk\.16\.attn_output\.weight$=q4_k
+^blk\.16\.attn_q\.weight$=q2_k
+^blk\.16\.attn_v\.weight$=q5_k
+^blk\.16\.ffn_down\.weight$=q2_k
+^blk\.16\.ffn_gate\.weight$=q2_k
+^blk\.16\.ffn_up\.weight$=q2_k
+^blk\.17\.attn_k\.weight$=q4_k
+^blk\.17\.attn_output\.weight$=q4_k
+^blk\.17\.attn_q\.weight$=q2_k
+^blk\.17\.attn_v\.weight$=q5_k
+^blk\.17\.ffn_down\.weight$=q2_k
+^blk\.17\.ffn_gate\.weight$=q2_k
+^blk\.17\.ffn_up\.weight$=q2_k
+^blk\.18\.attn_k\.weight$=q4_k
+^blk\.18\.attn_output\.weight$=q4_k
+^blk\.18\.attn_q\.weight$=q2_k
+^blk\.18\.attn_v\.weight$=q5_k
+^blk\.18\.ffn_down\.weight$=q2_k
+^blk\.18\.ffn_gate\.weight$=q2_k
+^blk\.18\.ffn_up\.weight$=q2_k
+^blk\.19\.attn_k\.weight$=q4_k
+^blk\.19\.attn_output\.weight$=q4_k
+^blk\.19\.attn_q\.weight$=q2_k
+^blk\.19\.attn_v\.weight$=q5_k
+^blk\.19\.ffn_down\.weight$=q2_k
+^blk\.19\.ffn_gate\.weight$=q2_k
+^blk\.19\.ffn_up\.weight$=q2_k
+^blk\.20\.attn_k\.weight$=q4_k
+^blk\.20\.attn_output\.weight$=q4_k
+^blk\.20\.attn_q\.weight$=q2_k
+^blk\.20\.attn_v\.weight$=q5_k
+^blk\.20\.ffn_down\.weight$=q2_k
+^blk\.20\.ffn_gate\.weight$=q2_k
+^blk\.20\.ffn_up\.weight$=q2_k
+^blk\.21\.attn_k\.weight$=q4_k
+^blk\.21\.attn_output\.weight$=q4_k
+^blk\.21\.attn_q\.weight$=q2_k
+^blk\.21\.attn_v\.weight$=q5_k
+^blk\.21\.ffn_down\.weight$=q2_k
+^blk\.21\.ffn_gate\.weight$=q2_k
+^blk\.21\.ffn_up\.weight$=q2_k
+^blk\.22\.attn_k\.weight$=q4_k
+^blk\.22\.attn_output\.weight$=q4_k
+^blk\.22\.attn_q\.weight$=q2_k
+^blk\.22\.attn_v\.weight$=q5_k
+^blk\.22\.ffn_down\.weight$=q2_k
+^blk\.22\.ffn_gate\.weight$=q2_k
+^blk\.22\.ffn_up\.weight$=q2_k
+^blk\.23\.attn_k\.weight$=q4_k
+^blk\.23\.attn_output\.weight$=q4_k
+^blk\.23\.attn_q\.weight$=q2_k
+^blk\.23\.attn_v\.weight$=q5_k
+^blk\.23\.ffn_down\.weight$=q2_k
+^blk\.23\.ffn_gate\.weight$=q2_k
+^blk\.23\.ffn_up\.weight$=q2_k
+^blk\.24\.attn_k\.weight$=q4_k
+^blk\.24\.attn_output\.weight$=q4_k
+^blk\.24\.attn_q\.weight$=q2_k
+^blk\.24\.attn_v\.weight$=q5_k
+^blk\.24\.ffn_down\.weight$=q2_k
+^blk\.24\.ffn_gate\.weight$=q2_k
+^blk\.24\.ffn_up\.weight$=q2_k
+^blk\.25\.attn_k\.weight$=q4_k
+^blk\.25\.attn_output\.weight$=q4_k
+^blk\.25\.attn_q\.weight$=q2_k
+^blk\.25\.attn_v\.weight$=q5_k
+^blk\.25\.ffn_down\.weight$=q2_k
+^blk\.25\.ffn_gate\.weight$=q2_k
+^blk\.25\.ffn_up\.weight$=q2_k
+^blk\.26\.attn_k\.weight$=q4_k
+^blk\.26\.attn_output\.weight$=q4_k
+^blk\.26\.attn_q\.weight$=q2_k
+^blk\.26\.attn_v\.weight$=q5_k
+^blk\.26\.ffn_down\.weight$=q2_k
+^blk\.26\.ffn_gate\.weight$=q2_k
+^blk\.26\.ffn_up\.weight$=q2_k
+^blk\.27\.attn_k\.weight$=q4_k
+^blk\.27\.attn_output\.weight$=q4_k
+^blk\.27\.attn_q\.weight$=q2_k
+^blk\.27\.attn_v\.weight$=q5_k
+^blk\.27\.ffn_down\.weight$=q2_k
+^blk\.27\.ffn_gate\.weight$=q2_k
+^blk\.27\.ffn_up\.weight$=q2_k
+^blk\.28\.attn_k\.weight$=q4_k
+^blk\.28\.attn_output\.weight$=q4_k
+^blk\.28\.attn_q\.weight$=q2_k
+^blk\.28\.attn_v\.weight$=q5_k
+^blk\.28\.ffn_down\.weight$=q2_k
+^blk\.28\.ffn_gate\.weight$=q2_k
+^blk\.28\.ffn_up\.weight$=q2_k
+^blk\.29\.attn_k\.weight$=q4_k
+^blk\.29\.attn_output\.weight$=q4_k
+^blk\.29\.attn_q\.weight$=q2_k
+^blk\.29\.attn_v\.weight$=q5_k
+^blk\.29\.ffn_down\.weight$=q2_k
+^blk\.29\.ffn_gate\.weight$=q2_k
+^blk\.29\.ffn_up\.weight$=q2_k
+^blk\.30\.attn_k\.weight$=q4_k
+^blk\.30\.attn_output\.weight$=q2_k
+^blk\.30\.attn_q\.weight$=q2_k
+^blk\.30\.attn_v\.weight$=q4_k
+^blk\.30\.ffn_down\.weight$=q2_k
+^blk\.30\.ffn_gate\.weight$=q2_k
+^blk\.30\.ffn_up\.weight$=q2_k
+^blk\.31\.attn_k\.weight$=q4_k
+^blk\.31\.attn_output\.weight$=q2_k
+^blk\.31\.attn_q\.weight$=q2_k
+^blk\.31\.attn_v\.weight$=q4_k
+^blk\.31\.ffn_down\.weight$=q2_k
+^blk\.31\.ffn_gate\.weight$=q2_k
+^blk\.31\.ffn_up\.weight$=q2_k
+^blk\.32\.attn_k\.weight$=q4_k
+^blk\.32\.attn_output\.weight$=q2_k
+^blk\.32\.attn_q\.weight$=q2_k
+^blk\.32\.attn_v\.weight$=q4_k
+^blk\.32\.ffn_down\.weight$=q2_k
+^blk\.32\.ffn_gate\.weight$=q2_k
+^blk\.32\.ffn_up\.weight$=q2_k
+^blk\.33\.attn_k\.weight$=q4_k
+^blk\.33\.attn_output\.weight$=q2_k
+^blk\.33\.attn_q\.weight$=q2_k
+^blk\.33\.attn_v\.weight$=q4_k
+^blk\.33\.ffn_down\.weight$=q2_k
+^blk\.33\.ffn_gate\.weight$=q2_k
+^blk\.33\.ffn_up\.weight$=q2_k
+^blk\.34\.attn_k\.weight$=q4_k
+^blk\.34\.attn_output\.weight$=q2_k
+^blk\.34\.attn_q\.weight$=q2_k
+^blk\.34\.attn_v\.weight$=q4_k
+^blk\.34\.ffn_down\.weight$=q2_k
+^blk\.34\.ffn_gate\.weight$=q2_k
+^blk\.34\.ffn_up\.weight$=q2_k
+^blk\.35\.attn_k\.weight$=q4_k
+^blk\.35\.attn_output\.weight$=q2_k
+^blk\.35\.attn_q\.weight$=q2_k
+^blk\.35\.attn_v\.weight$=q4_k
+^blk\.35\.ffn_down\.weight$=q2_k
+^blk\.35\.ffn_gate\.weight$=q2_k
+^blk\.35\.ffn_up\.weight$=q2_k
+^blk\.36\.attn_k\.weight$=q4_k
+^blk\.36\.attn_output\.weight$=q2_k
+^blk\.36\.attn_q\.weight$=q2_k
+^blk\.36\.attn_v\.weight$=q4_k
+^blk\.36\.ffn_down\.weight$=q2_k
+^blk\.36\.ffn_gate\.weight$=q2_k
+^blk\.36\.ffn_up\.weight$=q2_k
+^blk\.37\.attn_k\.weight$=q4_k
+^blk\.37\.attn_output\.weight$=q2_k
+^blk\.37\.attn_q\.weight$=q2_k
+^blk\.37\.attn_v\.weight$=q4_k
+^blk\.37\.ffn_down\.weight$=q2_k
+^blk\.37\.ffn_gate\.weight$=q2_k
+^blk\.37\.ffn_up\.weight$=q2_k
+^blk\.38\.attn_k\.weight$=q4_k
+^blk\.38\.attn_output\.weight$=q2_k
+^blk\.38\.attn_q\.weight$=q2_k
+^blk\.38\.attn_v\.weight$=q4_k
+^blk\.38\.ffn_down\.weight$=q2_k
+^blk\.38\.ffn_gate\.weight$=q2_k
+^blk\.38\.ffn_up\.weight$=q2_k
+^blk\.39\.attn_k\.weight$=q4_k
+^blk\.39\.attn_output\.weight$=q2_k
+^blk\.39\.attn_q\.weight$=q2_k
+^blk\.39\.attn_v\.weight$=q4_k
+^blk\.39\.ffn_down\.weight$=q4_k
+^blk\.39\.ffn_gate\.weight$=q2_k
+^blk\.39\.ffn_up\.weight$=q4_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_L.layout.json

@@ -0,0 +1,88 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q3_K_L",
+  "base_type": "q3_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.50",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "L",
+  "rung_share": 0.5,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.08,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0419,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 7965024000,
+    "map_body+embd": 7566524160,
+    "ratio": 0.95,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.5,
+  "body_bpw": 3.994,
+  "body_bpw_heuristic": 4.235,
+  "total_params": 14765593600,
+  "file_bytes_pred": 7568224000,
+  "file_bpw_pred": 4.1,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 40
+    },
+    "attn_output": {
+      "q5_k": 30,
+      "q3_k": 9,
+      "q4_k": 1
+    },
+    "attn_q": {
+      "q3_k": 39,
+      "q4_k": 1
+    },
+    "attn_v": {
+      "q6_k": 31,
+      "q5_k": 9
+    },
+    "ffn_down": {
+      "q4_k": 20,
+      "q3_k": 20
+    },
+    "ffn_gate": {
+      "q3_k": 32,
+      "q4_k": 8
+    },
+    "ffn_up": {
+      "q3_k": 20,
+      "q4_k": 19,
+      "q5_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q3_K_L",
+  "quantize_ftype": "Q3_K_L",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_L.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q5_k
+^blk\.0\.attn_output\.weight$=q5_k
+^blk\.0\.attn_q\.weight$=q3_k
+^blk\.0\.attn_v\.weight$=q6_k
+^blk\.0\.ffn_down\.weight$=q4_k
+^blk\.0\.ffn_gate\.weight$=q3_k
+^blk\.0\.ffn_up\.weight$=q4_k
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=q3_k
+^blk\.1\.attn_v\.weight$=q6_k
+^blk\.1\.ffn_down\.weight$=q4_k
+^blk\.1\.ffn_gate\.weight$=q4_k
+^blk\.1\.ffn_up\.weight$=q4_k
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=q3_k
+^blk\.2\.attn_v\.weight$=q6_k
+^blk\.2\.ffn_down\.weight$=q4_k
+^blk\.2\.ffn_gate\.weight$=q4_k
+^blk\.2\.ffn_up\.weight$=q4_k
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=q3_k
+^blk\.3\.attn_v\.weight$=q6_k
+^blk\.3\.ffn_down\.weight$=q4_k
+^blk\.3\.ffn_gate\.weight$=q4_k
+^blk\.3\.ffn_up\.weight$=q4_k
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=q3_k
+^blk\.4\.attn_v\.weight$=q6_k
+^blk\.4\.ffn_down\.weight$=q4_k
+^blk\.4\.ffn_gate\.weight$=q4_k
+^blk\.4\.ffn_up\.weight$=q4_k
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=q3_k
+^blk\.5\.attn_v\.weight$=q6_k
+^blk\.5\.ffn_down\.weight$=q4_k
+^blk\.5\.ffn_gate\.weight$=q4_k
+^blk\.5\.ffn_up\.weight$=q4_k
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=q3_k
+^blk\.6\.attn_v\.weight$=q6_k
+^blk\.6\.ffn_down\.weight$=q4_k
+^blk\.6\.ffn_gate\.weight$=q4_k
+^blk\.6\.ffn_up\.weight$=q4_k
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=q3_k
+^blk\.7\.attn_v\.weight$=q6_k
+^blk\.7\.ffn_down\.weight$=q4_k
+^blk\.7\.ffn_gate\.weight$=q4_k
+^blk\.7\.ffn_up\.weight$=q4_k
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=q3_k
+^blk\.8\.attn_v\.weight$=q6_k
+^blk\.8\.ffn_down\.weight$=q4_k
+^blk\.8\.ffn_gate\.weight$=q3_k
+^blk\.8\.ffn_up\.weight$=q4_k
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=q3_k
+^blk\.9\.attn_v\.weight$=q6_k
+^blk\.9\.ffn_down\.weight$=q4_k
+^blk\.9\.ffn_gate\.weight$=q3_k
+^blk\.9\.ffn_up\.weight$=q4_k
+^blk\.10\.attn_k\.weight$=q5_k
+^blk\.10\.attn_output\.weight$=q5_k
+^blk\.10\.attn_q\.weight$=q4_k
+^blk\.10\.attn_v\.weight$=q6_k
+^blk\.10\.ffn_down\.weight$=q3_k
+^blk\.10\.ffn_gate\.weight$=q3_k
+^blk\.10\.ffn_up\.weight$=q3_k
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=q5_k
+^blk\.11\.attn_q\.weight$=q3_k
+^blk\.11\.attn_v\.weight$=q6_k
+^blk\.11\.ffn_down\.weight$=q3_k
+^blk\.11\.ffn_gate\.weight$=q3_k
+^blk\.11\.ffn_up\.weight$=q3_k
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=q5_k
+^blk\.12\.attn_q\.weight$=q3_k
+^blk\.12\.attn_v\.weight$=q6_k
+^blk\.12\.ffn_down\.weight$=q3_k
+^blk\.12\.ffn_gate\.weight$=q3_k
+^blk\.12\.ffn_up\.weight$=q3_k
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=q5_k
+^blk\.13\.attn_q\.weight$=q3_k
+^blk\.13\.attn_v\.weight$=q6_k
+^blk\.13\.ffn_down\.weight$=q3_k
+^blk\.13\.ffn_gate\.weight$=q3_k
+^blk\.13\.ffn_up\.weight$=q3_k
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=q5_k
+^blk\.14\.attn_q\.weight$=q3_k
+^blk\.14\.attn_v\.weight$=q6_k
+^blk\.14\.ffn_down\.weight$=q3_k
+^blk\.14\.ffn_gate\.weight$=q3_k
+^blk\.14\.ffn_up\.weight$=q3_k
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=q5_k
+^blk\.15\.attn_q\.weight$=q3_k
+^blk\.15\.attn_v\.weight$=q6_k
+^blk\.15\.ffn_down\.weight$=q3_k
+^blk\.15\.ffn_gate\.weight$=q3_k
+^blk\.15\.ffn_up\.weight$=q3_k
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=q5_k
+^blk\.16\.attn_q\.weight$=q3_k
+^blk\.16\.attn_v\.weight$=q6_k
+^blk\.16\.ffn_down\.weight$=q3_k
+^blk\.16\.ffn_gate\.weight$=q3_k
+^blk\.16\.ffn_up\.weight$=q3_k
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=q5_k
+^blk\.17\.attn_q\.weight$=q3_k
+^blk\.17\.attn_v\.weight$=q6_k
+^blk\.17\.ffn_down\.weight$=q3_k
+^blk\.17\.ffn_gate\.weight$=q3_k
+^blk\.17\.ffn_up\.weight$=q3_k
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=q5_k
+^blk\.18\.attn_q\.weight$=q3_k
+^blk\.18\.attn_v\.weight$=q6_k
+^blk\.18\.ffn_down\.weight$=q3_k
+^blk\.18\.ffn_gate\.weight$=q3_k
+^blk\.18\.ffn_up\.weight$=q3_k
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=q5_k
+^blk\.19\.attn_q\.weight$=q3_k
+^blk\.19\.attn_v\.weight$=q6_k
+^blk\.19\.ffn_down\.weight$=q3_k
+^blk\.19\.ffn_gate\.weight$=q3_k
+^blk\.19\.ffn_up\.weight$=q3_k
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=q5_k
+^blk\.20\.attn_q\.weight$=q3_k
+^blk\.20\.attn_v\.weight$=q6_k
+^blk\.20\.ffn_down\.weight$=q3_k
+^blk\.20\.ffn_gate\.weight$=q3_k
+^blk\.20\.ffn_up\.weight$=q3_k
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=q5_k
+^blk\.21\.attn_q\.weight$=q3_k
+^blk\.21\.attn_v\.weight$=q6_k
+^blk\.21\.ffn_down\.weight$=q3_k
+^blk\.21\.ffn_gate\.weight$=q3_k
+^blk\.21\.ffn_up\.weight$=q3_k
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=q5_k
+^blk\.22\.attn_q\.weight$=q3_k
+^blk\.22\.attn_v\.weight$=q6_k
+^blk\.22\.ffn_down\.weight$=q3_k
+^blk\.22\.ffn_gate\.weight$=q3_k
+^blk\.22\.ffn_up\.weight$=q3_k
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=q5_k
+^blk\.23\.attn_q\.weight$=q3_k
+^blk\.23\.attn_v\.weight$=q6_k
+^blk\.23\.ffn_down\.weight$=q3_k
+^blk\.23\.ffn_gate\.weight$=q3_k
+^blk\.23\.ffn_up\.weight$=q3_k
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=q5_k
+^blk\.24\.attn_q\.weight$=q3_k
+^blk\.24\.attn_v\.weight$=q6_k
+^blk\.24\.ffn_down\.weight$=q3_k
+^blk\.24\.ffn_gate\.weight$=q3_k
+^blk\.24\.ffn_up\.weight$=q3_k
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=q5_k
+^blk\.25\.attn_q\.weight$=q3_k
+^blk\.25\.attn_v\.weight$=q6_k
+^blk\.25\.ffn_down\.weight$=q3_k
+^blk\.25\.ffn_gate\.weight$=q3_k
+^blk\.25\.ffn_up\.weight$=q3_k
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=q5_k
+^blk\.26\.attn_q\.weight$=q3_k
+^blk\.26\.attn_v\.weight$=q6_k
+^blk\.26\.ffn_down\.weight$=q3_k
+^blk\.26\.ffn_gate\.weight$=q3_k
+^blk\.26\.ffn_up\.weight$=q3_k
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=q5_k
+^blk\.27\.attn_q\.weight$=q3_k
+^blk\.27\.attn_v\.weight$=q6_k
+^blk\.27\.ffn_down\.weight$=q3_k
+^blk\.27\.ffn_gate\.weight$=q3_k
+^blk\.27\.ffn_up\.weight$=q3_k
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=q5_k
+^blk\.28\.attn_q\.weight$=q3_k
+^blk\.28\.attn_v\.weight$=q6_k
+^blk\.28\.ffn_down\.weight$=q3_k
+^blk\.28\.ffn_gate\.weight$=q3_k
+^blk\.28\.ffn_up\.weight$=q3_k
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=q5_k
+^blk\.29\.attn_q\.weight$=q3_k
+^blk\.29\.attn_v\.weight$=q6_k
+^blk\.29\.ffn_down\.weight$=q3_k
+^blk\.29\.ffn_gate\.weight$=q3_k
+^blk\.29\.ffn_up\.weight$=q3_k
+^blk\.30\.attn_k\.weight$=q5_k
+^blk\.30\.attn_output\.weight$=q3_k
+^blk\.30\.attn_q\.weight$=q3_k
+^blk\.30\.attn_v\.weight$=q5_k
+^blk\.30\.ffn_down\.weight$=q4_k
+^blk\.30\.ffn_gate\.weight$=q3_k
+^blk\.30\.ffn_up\.weight$=q4_k
+^blk\.31\.attn_k\.weight$=q5_k
+^blk\.31\.attn_output\.weight$=q3_k
+^blk\.31\.attn_q\.weight$=q3_k
+^blk\.31\.attn_v\.weight$=q5_k
+^blk\.31\.ffn_down\.weight$=q4_k
+^blk\.31\.ffn_gate\.weight$=q3_k
+^blk\.31\.ffn_up\.weight$=q4_k
+^blk\.32\.attn_k\.weight$=q5_k
+^blk\.32\.attn_output\.weight$=q3_k
+^blk\.32\.attn_q\.weight$=q3_k
+^blk\.32\.attn_v\.weight$=q5_k
+^blk\.32\.ffn_down\.weight$=q4_k
+^blk\.32\.ffn_gate\.weight$=q3_k
+^blk\.32\.ffn_up\.weight$=q4_k
+^blk\.33\.attn_k\.weight$=q5_k
+^blk\.33\.attn_output\.weight$=q3_k
+^blk\.33\.attn_q\.weight$=q3_k
+^blk\.33\.attn_v\.weight$=q5_k
+^blk\.33\.ffn_down\.weight$=q4_k
+^blk\.33\.ffn_gate\.weight$=q3_k
+^blk\.33\.ffn_up\.weight$=q4_k
+^blk\.34\.attn_k\.weight$=q5_k
+^blk\.34\.attn_output\.weight$=q3_k
+^blk\.34\.attn_q\.weight$=q3_k
+^blk\.34\.attn_v\.weight$=q5_k
+^blk\.34\.ffn_down\.weight$=q4_k
+^blk\.34\.ffn_gate\.weight$=q3_k
+^blk\.34\.ffn_up\.weight$=q4_k
+^blk\.35\.attn_k\.weight$=q5_k
+^blk\.35\.attn_output\.weight$=q3_k
+^blk\.35\.attn_q\.weight$=q3_k
+^blk\.35\.attn_v\.weight$=q5_k
+^blk\.35\.ffn_down\.weight$=q4_k
+^blk\.35\.ffn_gate\.weight$=q3_k
+^blk\.35\.ffn_up\.weight$=q4_k
+^blk\.36\.attn_k\.weight$=q5_k
+^blk\.36\.attn_output\.weight$=q3_k
+^blk\.36\.attn_q\.weight$=q3_k
+^blk\.36\.attn_v\.weight$=q5_k
+^blk\.36\.ffn_down\.weight$=q4_k
+^blk\.36\.ffn_gate\.weight$=q3_k
+^blk\.36\.ffn_up\.weight$=q4_k
+^blk\.37\.attn_k\.weight$=q5_k
+^blk\.37\.attn_output\.weight$=q3_k
+^blk\.37\.attn_q\.weight$=q3_k
+^blk\.37\.attn_v\.weight$=q5_k
+^blk\.37\.ffn_down\.weight$=q4_k
+^blk\.37\.ffn_gate\.weight$=q3_k
+^blk\.37\.ffn_up\.weight$=q4_k
+^blk\.38\.attn_k\.weight$=q5_k
+^blk\.38\.attn_output\.weight$=q3_k
+^blk\.38\.attn_q\.weight$=q3_k
+^blk\.38\.attn_v\.weight$=q5_k
+^blk\.38\.ffn_down\.weight$=q4_k
+^blk\.38\.ffn_gate\.weight$=q3_k
+^blk\.38\.ffn_up\.weight$=q4_k
+^blk\.39\.attn_k\.weight$=q5_k
+^blk\.39\.attn_output\.weight$=q4_k
+^blk\.39\.attn_q\.weight$=q3_k
+^blk\.39\.attn_v\.weight$=q6_k
+^blk\.39\.ffn_down\.weight$=q4_k
+^blk\.39\.ffn_gate\.weight$=q4_k
+^blk\.39\.ffn_up\.weight$=q5_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_M.layout.json

@@ -0,0 +1,87 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q3_K_M",
+  "base_type": "q3_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0846,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0443,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 7528554240,
+    "map_body+embd": 7159013120,
+    "ratio": 0.9509,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 3.747,
+  "body_bpw_heuristic": 3.971,
+  "total_params": 14765593600,
+  "file_bytes_pred": 7160712960,
+  "file_bpw_pred": 3.88,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 30,
+      "q4_k": 10
+    },
+    "attn_output": {
+      "q4_k": 22,
+      "q5_k": 9,
+      "q3_k": 9
+    },
+    "attn_q": {
+      "q3_k": 39,
+      "q4_k": 1
+    },
+    "attn_v": {
+      "q6_k": 29,
+      "q5_k": 11
+    },
+    "ffn_down": {
+      "q3_k": 29,
+      "q4_k": 11
+    },
+    "ffn_gate": {
+      "q3_k": 40
+    },
+    "ffn_up": {
+      "q3_k": 30,
+      "q4_k": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q3_K_M",
+  "quantize_ftype": "Q3_K_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_M.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q5_k
+^blk\.0\.attn_output\.weight$=q4_k
+^blk\.0\.attn_q\.weight$=q3_k
+^blk\.0\.attn_v\.weight$=q5_k
+^blk\.0\.ffn_down\.weight$=q4_k
+^blk\.0\.ffn_gate\.weight$=q3_k
+^blk\.0\.ffn_up\.weight$=q3_k
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=q3_k
+^blk\.1\.attn_v\.weight$=q6_k
+^blk\.1\.ffn_down\.weight$=q4_k
+^blk\.1\.ffn_gate\.weight$=q3_k
+^blk\.1\.ffn_up\.weight$=q4_k
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=q3_k
+^blk\.2\.attn_v\.weight$=q6_k
+^blk\.2\.ffn_down\.weight$=q4_k
+^blk\.2\.ffn_gate\.weight$=q3_k
+^blk\.2\.ffn_up\.weight$=q4_k
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=q3_k
+^blk\.3\.attn_v\.weight$=q6_k
+^blk\.3\.ffn_down\.weight$=q4_k
+^blk\.3\.ffn_gate\.weight$=q3_k
+^blk\.3\.ffn_up\.weight$=q4_k
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=q3_k
+^blk\.4\.attn_v\.weight$=q6_k
+^blk\.4\.ffn_down\.weight$=q4_k
+^blk\.4\.ffn_gate\.weight$=q3_k
+^blk\.4\.ffn_up\.weight$=q4_k
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=q3_k
+^blk\.5\.attn_v\.weight$=q6_k
+^blk\.5\.ffn_down\.weight$=q4_k
+^blk\.5\.ffn_gate\.weight$=q3_k
+^blk\.5\.ffn_up\.weight$=q4_k
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=q3_k
+^blk\.6\.attn_v\.weight$=q6_k
+^blk\.6\.ffn_down\.weight$=q4_k
+^blk\.6\.ffn_gate\.weight$=q3_k
+^blk\.6\.ffn_up\.weight$=q4_k
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=q3_k
+^blk\.7\.attn_v\.weight$=q6_k
+^blk\.7\.ffn_down\.weight$=q4_k
+^blk\.7\.ffn_gate\.weight$=q3_k
+^blk\.7\.ffn_up\.weight$=q4_k
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=q3_k
+^blk\.8\.attn_v\.weight$=q6_k
+^blk\.8\.ffn_down\.weight$=q4_k
+^blk\.8\.ffn_gate\.weight$=q3_k
+^blk\.8\.ffn_up\.weight$=q4_k
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=q3_k
+^blk\.9\.attn_v\.weight$=q6_k
+^blk\.9\.ffn_down\.weight$=q4_k
+^blk\.9\.ffn_gate\.weight$=q3_k
+^blk\.9\.ffn_up\.weight$=q4_k
+^blk\.10\.attn_k\.weight$=q5_k
+^blk\.10\.attn_output\.weight$=q4_k
+^blk\.10\.attn_q\.weight$=q4_k
+^blk\.10\.attn_v\.weight$=q6_k
+^blk\.10\.ffn_down\.weight$=q3_k
+^blk\.10\.ffn_gate\.weight$=q3_k
+^blk\.10\.ffn_up\.weight$=q3_k
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=q4_k
+^blk\.11\.attn_q\.weight$=q3_k
+^blk\.11\.attn_v\.weight$=q6_k
+^blk\.11\.ffn_down\.weight$=q3_k
+^blk\.11\.ffn_gate\.weight$=q3_k
+^blk\.11\.ffn_up\.weight$=q3_k
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=q4_k
+^blk\.12\.attn_q\.weight$=q3_k
+^blk\.12\.attn_v\.weight$=q6_k
+^blk\.12\.ffn_down\.weight$=q3_k
+^blk\.12\.ffn_gate\.weight$=q3_k
+^blk\.12\.ffn_up\.weight$=q3_k
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=q4_k
+^blk\.13\.attn_q\.weight$=q3_k
+^blk\.13\.attn_v\.weight$=q6_k
+^blk\.13\.ffn_down\.weight$=q3_k
+^blk\.13\.ffn_gate\.weight$=q3_k
+^blk\.13\.ffn_up\.weight$=q3_k
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=q4_k
+^blk\.14\.attn_q\.weight$=q3_k
+^blk\.14\.attn_v\.weight$=q6_k
+^blk\.14\.ffn_down\.weight$=q3_k
+^blk\.14\.ffn_gate\.weight$=q3_k
+^blk\.14\.ffn_up\.weight$=q3_k
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=q4_k
+^blk\.15\.attn_q\.weight$=q3_k
+^blk\.15\.attn_v\.weight$=q6_k
+^blk\.15\.ffn_down\.weight$=q3_k
+^blk\.15\.ffn_gate\.weight$=q3_k
+^blk\.15\.ffn_up\.weight$=q3_k
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=q4_k
+^blk\.16\.attn_q\.weight$=q3_k
+^blk\.16\.attn_v\.weight$=q6_k
+^blk\.16\.ffn_down\.weight$=q3_k
+^blk\.16\.ffn_gate\.weight$=q3_k
+^blk\.16\.ffn_up\.weight$=q3_k
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=q4_k
+^blk\.17\.attn_q\.weight$=q3_k
+^blk\.17\.attn_v\.weight$=q6_k
+^blk\.17\.ffn_down\.weight$=q3_k
+^blk\.17\.ffn_gate\.weight$=q3_k
+^blk\.17\.ffn_up\.weight$=q3_k
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=q4_k
+^blk\.18\.attn_q\.weight$=q3_k
+^blk\.18\.attn_v\.weight$=q6_k
+^blk\.18\.ffn_down\.weight$=q3_k
+^blk\.18\.ffn_gate\.weight$=q3_k
+^blk\.18\.ffn_up\.weight$=q3_k
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=q4_k
+^blk\.19\.attn_q\.weight$=q3_k
+^blk\.19\.attn_v\.weight$=q6_k
+^blk\.19\.ffn_down\.weight$=q3_k
+^blk\.19\.ffn_gate\.weight$=q3_k
+^blk\.19\.ffn_up\.weight$=q3_k
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=q4_k
+^blk\.20\.attn_q\.weight$=q3_k
+^blk\.20\.attn_v\.weight$=q6_k
+^blk\.20\.ffn_down\.weight$=q3_k
+^blk\.20\.ffn_gate\.weight$=q3_k
+^blk\.20\.ffn_up\.weight$=q3_k
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=q4_k
+^blk\.21\.attn_q\.weight$=q3_k
+^blk\.21\.attn_v\.weight$=q6_k
+^blk\.21\.ffn_down\.weight$=q3_k
+^blk\.21\.ffn_gate\.weight$=q3_k
+^blk\.21\.ffn_up\.weight$=q3_k
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=q4_k
+^blk\.22\.attn_q\.weight$=q3_k
+^blk\.22\.attn_v\.weight$=q6_k
+^blk\.22\.ffn_down\.weight$=q3_k
+^blk\.22\.ffn_gate\.weight$=q3_k
+^blk\.22\.ffn_up\.weight$=q3_k
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=q4_k
+^blk\.23\.attn_q\.weight$=q3_k
+^blk\.23\.attn_v\.weight$=q6_k
+^blk\.23\.ffn_down\.weight$=q3_k
+^blk\.23\.ffn_gate\.weight$=q3_k
+^blk\.23\.ffn_up\.weight$=q3_k
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=q4_k
+^blk\.24\.attn_q\.weight$=q3_k
+^blk\.24\.attn_v\.weight$=q6_k
+^blk\.24\.ffn_down\.weight$=q3_k
+^blk\.24\.ffn_gate\.weight$=q3_k
+^blk\.24\.ffn_up\.weight$=q3_k
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=q4_k
+^blk\.25\.attn_q\.weight$=q3_k
+^blk\.25\.attn_v\.weight$=q6_k
+^blk\.25\.ffn_down\.weight$=q3_k
+^blk\.25\.ffn_gate\.weight$=q3_k
+^blk\.25\.ffn_up\.weight$=q3_k
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=q4_k
+^blk\.26\.attn_q\.weight$=q3_k
+^blk\.26\.attn_v\.weight$=q6_k
+^blk\.26\.ffn_down\.weight$=q3_k
+^blk\.26\.ffn_gate\.weight$=q3_k
+^blk\.26\.ffn_up\.weight$=q3_k
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=q4_k
+^blk\.27\.attn_q\.weight$=q3_k
+^blk\.27\.attn_v\.weight$=q6_k
+^blk\.27\.ffn_down\.weight$=q3_k
+^blk\.27\.ffn_gate\.weight$=q3_k
+^blk\.27\.ffn_up\.weight$=q3_k
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=q4_k
+^blk\.28\.attn_q\.weight$=q3_k
+^blk\.28\.attn_v\.weight$=q6_k
+^blk\.28\.ffn_down\.weight$=q3_k
+^blk\.28\.ffn_gate\.weight$=q3_k
+^blk\.28\.ffn_up\.weight$=q3_k
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=q4_k
+^blk\.29\.attn_q\.weight$=q3_k
+^blk\.29\.attn_v\.weight$=q6_k
+^blk\.29\.ffn_down\.weight$=q3_k
+^blk\.29\.ffn_gate\.weight$=q3_k
+^blk\.29\.ffn_up\.weight$=q3_k
+^blk\.30\.attn_k\.weight$=q4_k
+^blk\.30\.attn_output\.weight$=q3_k
+^blk\.30\.attn_q\.weight$=q3_k
+^blk\.30\.attn_v\.weight$=q5_k
+^blk\.30\.ffn_down\.weight$=q3_k
+^blk\.30\.ffn_gate\.weight$=q3_k
+^blk\.30\.ffn_up\.weight$=q3_k
+^blk\.31\.attn_k\.weight$=q4_k
+^blk\.31\.attn_output\.weight$=q3_k
+^blk\.31\.attn_q\.weight$=q3_k
+^blk\.31\.attn_v\.weight$=q5_k
+^blk\.31\.ffn_down\.weight$=q3_k
+^blk\.31\.ffn_gate\.weight$=q3_k
+^blk\.31\.ffn_up\.weight$=q3_k
+^blk\.32\.attn_k\.weight$=q4_k
+^blk\.32\.attn_output\.weight$=q3_k
+^blk\.32\.attn_q\.weight$=q3_k
+^blk\.32\.attn_v\.weight$=q5_k
+^blk\.32\.ffn_down\.weight$=q3_k
+^blk\.32\.ffn_gate\.weight$=q3_k
+^blk\.32\.ffn_up\.weight$=q3_k
+^blk\.33\.attn_k\.weight$=q4_k
+^blk\.33\.attn_output\.weight$=q3_k
+^blk\.33\.attn_q\.weight$=q3_k
+^blk\.33\.attn_v\.weight$=q5_k
+^blk\.33\.ffn_down\.weight$=q3_k
+^blk\.33\.ffn_gate\.weight$=q3_k
+^blk\.33\.ffn_up\.weight$=q3_k
+^blk\.34\.attn_k\.weight$=q4_k
+^blk\.34\.attn_output\.weight$=q3_k
+^blk\.34\.attn_q\.weight$=q3_k
+^blk\.34\.attn_v\.weight$=q5_k
+^blk\.34\.ffn_down\.weight$=q3_k
+^blk\.34\.ffn_gate\.weight$=q3_k
+^blk\.34\.ffn_up\.weight$=q3_k
+^blk\.35\.attn_k\.weight$=q4_k
+^blk\.35\.attn_output\.weight$=q3_k
+^blk\.35\.attn_q\.weight$=q3_k
+^blk\.35\.attn_v\.weight$=q5_k
+^blk\.35\.ffn_down\.weight$=q3_k
+^blk\.35\.ffn_gate\.weight$=q3_k
+^blk\.35\.ffn_up\.weight$=q3_k
+^blk\.36\.attn_k\.weight$=q4_k
+^blk\.36\.attn_output\.weight$=q3_k
+^blk\.36\.attn_q\.weight$=q3_k
+^blk\.36\.attn_v\.weight$=q5_k
+^blk\.36\.ffn_down\.weight$=q3_k
+^blk\.36\.ffn_gate\.weight$=q3_k
+^blk\.36\.ffn_up\.weight$=q3_k
+^blk\.37\.attn_k\.weight$=q4_k
+^blk\.37\.attn_output\.weight$=q3_k
+^blk\.37\.attn_q\.weight$=q3_k
+^blk\.37\.attn_v\.weight$=q5_k
+^blk\.37\.ffn_down\.weight$=q3_k
+^blk\.37\.ffn_gate\.weight$=q3_k
+^blk\.37\.ffn_up\.weight$=q3_k
+^blk\.38\.attn_k\.weight$=q4_k
+^blk\.38\.attn_output\.weight$=q3_k
+^blk\.38\.attn_q\.weight$=q3_k
+^blk\.38\.attn_v\.weight$=q5_k
+^blk\.38\.ffn_down\.weight$=q3_k
+^blk\.38\.ffn_gate\.weight$=q3_k
+^blk\.38\.ffn_up\.weight$=q3_k
+^blk\.39\.attn_k\.weight$=q4_k
+^blk\.39\.attn_output\.weight$=q4_k
+^blk\.39\.attn_q\.weight$=q3_k
+^blk\.39\.attn_v\.weight$=q5_k
+^blk\.39\.ffn_down\.weight$=q4_k
+^blk\.39\.ffn_gate\.weight$=q3_k
+^blk\.39\.ffn_up\.weight$=q4_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_S.layout.json

@@ -0,0 +1,85 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q3_K_S",
+  "base_type": "q3_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0929,
+      "type": "q5_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0487,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q5_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 6852550400,
+    "map_body+embd": 6860291840,
+    "ratio": 1.0011,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.9,
+  "body_bpw": 3.566,
+  "body_bpw_heuristic": 3.562,
+  "total_params": 14765593600,
+  "file_bytes_pred": 6861991680,
+  "file_bpw_pred": 3.718,
+  "histogram": {
+    "attn_k": {
+      "q5_k": 30,
+      "q4_k": 10
+    },
+    "attn_output": {
+      "q3_k": 26,
+      "q5_k": 9,
+      "q4_k": 5
+    },
+    "attn_q": {
+      "q3_k": 40
+    },
+    "attn_v": {
+      "q6_k": 30,
+      "q5_k": 10
+    },
+    "ffn_down": {
+      "q3_k": 40
+    },
+    "ffn_gate": {
+      "q3_k": 40
+    },
+    "ffn_up": {
+      "q3_k": 39,
+      "q4_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q3_K_S",
+  "quantize_ftype": "Q3_K_S",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q3_K_S.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q5_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q5_k
+^blk\.0\.attn_output\.weight$=q3_k
+^blk\.0\.attn_q\.weight$=q3_k
+^blk\.0\.attn_v\.weight$=q5_k
+^blk\.0\.ffn_down\.weight$=q3_k
+^blk\.0\.ffn_gate\.weight$=q3_k
+^blk\.0\.ffn_up\.weight$=q3_k
+^blk\.1\.attn_k\.weight$=q5_k
+^blk\.1\.attn_output\.weight$=q5_k
+^blk\.1\.attn_q\.weight$=q3_k
+^blk\.1\.attn_v\.weight$=q6_k
+^blk\.1\.ffn_down\.weight$=q3_k
+^blk\.1\.ffn_gate\.weight$=q3_k
+^blk\.1\.ffn_up\.weight$=q3_k
+^blk\.2\.attn_k\.weight$=q5_k
+^blk\.2\.attn_output\.weight$=q5_k
+^blk\.2\.attn_q\.weight$=q3_k
+^blk\.2\.attn_v\.weight$=q6_k
+^blk\.2\.ffn_down\.weight$=q3_k
+^blk\.2\.ffn_gate\.weight$=q3_k
+^blk\.2\.ffn_up\.weight$=q3_k
+^blk\.3\.attn_k\.weight$=q5_k
+^blk\.3\.attn_output\.weight$=q5_k
+^blk\.3\.attn_q\.weight$=q3_k
+^blk\.3\.attn_v\.weight$=q6_k
+^blk\.3\.ffn_down\.weight$=q3_k
+^blk\.3\.ffn_gate\.weight$=q3_k
+^blk\.3\.ffn_up\.weight$=q3_k
+^blk\.4\.attn_k\.weight$=q5_k
+^blk\.4\.attn_output\.weight$=q5_k
+^blk\.4\.attn_q\.weight$=q3_k
+^blk\.4\.attn_v\.weight$=q6_k
+^blk\.4\.ffn_down\.weight$=q3_k
+^blk\.4\.ffn_gate\.weight$=q3_k
+^blk\.4\.ffn_up\.weight$=q3_k
+^blk\.5\.attn_k\.weight$=q5_k
+^blk\.5\.attn_output\.weight$=q5_k
+^blk\.5\.attn_q\.weight$=q3_k
+^blk\.5\.attn_v\.weight$=q6_k
+^blk\.5\.ffn_down\.weight$=q3_k
+^blk\.5\.ffn_gate\.weight$=q3_k
+^blk\.5\.ffn_up\.weight$=q3_k
+^blk\.6\.attn_k\.weight$=q5_k
+^blk\.6\.attn_output\.weight$=q5_k
+^blk\.6\.attn_q\.weight$=q3_k
+^blk\.6\.attn_v\.weight$=q6_k
+^blk\.6\.ffn_down\.weight$=q3_k
+^blk\.6\.ffn_gate\.weight$=q3_k
+^blk\.6\.ffn_up\.weight$=q3_k
+^blk\.7\.attn_k\.weight$=q5_k
+^blk\.7\.attn_output\.weight$=q5_k
+^blk\.7\.attn_q\.weight$=q3_k
+^blk\.7\.attn_v\.weight$=q6_k
+^blk\.7\.ffn_down\.weight$=q3_k
+^blk\.7\.ffn_gate\.weight$=q3_k
+^blk\.7\.ffn_up\.weight$=q3_k
+^blk\.8\.attn_k\.weight$=q5_k
+^blk\.8\.attn_output\.weight$=q5_k
+^blk\.8\.attn_q\.weight$=q3_k
+^blk\.8\.attn_v\.weight$=q6_k
+^blk\.8\.ffn_down\.weight$=q3_k
+^blk\.8\.ffn_gate\.weight$=q3_k
+^blk\.8\.ffn_up\.weight$=q3_k
+^blk\.9\.attn_k\.weight$=q5_k
+^blk\.9\.attn_output\.weight$=q5_k
+^blk\.9\.attn_q\.weight$=q3_k
+^blk\.9\.attn_v\.weight$=q6_k
+^blk\.9\.ffn_down\.weight$=q3_k
+^blk\.9\.ffn_gate\.weight$=q3_k
+^blk\.9\.ffn_up\.weight$=q3_k
+^blk\.10\.attn_k\.weight$=q5_k
+^blk\.10\.attn_output\.weight$=q4_k
+^blk\.10\.attn_q\.weight$=q3_k
+^blk\.10\.attn_v\.weight$=q6_k
+^blk\.10\.ffn_down\.weight$=q3_k
+^blk\.10\.ffn_gate\.weight$=q3_k
+^blk\.10\.ffn_up\.weight$=q3_k
+^blk\.11\.attn_k\.weight$=q5_k
+^blk\.11\.attn_output\.weight$=q4_k
+^blk\.11\.attn_q\.weight$=q3_k
+^blk\.11\.attn_v\.weight$=q6_k
+^blk\.11\.ffn_down\.weight$=q3_k
+^blk\.11\.ffn_gate\.weight$=q3_k
+^blk\.11\.ffn_up\.weight$=q3_k
+^blk\.12\.attn_k\.weight$=q5_k
+^blk\.12\.attn_output\.weight$=q4_k
+^blk\.12\.attn_q\.weight$=q3_k
+^blk\.12\.attn_v\.weight$=q6_k
+^blk\.12\.ffn_down\.weight$=q3_k
+^blk\.12\.ffn_gate\.weight$=q3_k
+^blk\.12\.ffn_up\.weight$=q3_k
+^blk\.13\.attn_k\.weight$=q5_k
+^blk\.13\.attn_output\.weight$=q4_k
+^blk\.13\.attn_q\.weight$=q3_k
+^blk\.13\.attn_v\.weight$=q6_k
+^blk\.13\.ffn_down\.weight$=q3_k
+^blk\.13\.ffn_gate\.weight$=q3_k
+^blk\.13\.ffn_up\.weight$=q3_k
+^blk\.14\.attn_k\.weight$=q5_k
+^blk\.14\.attn_output\.weight$=q4_k
+^blk\.14\.attn_q\.weight$=q3_k
+^blk\.14\.attn_v\.weight$=q6_k
+^blk\.14\.ffn_down\.weight$=q3_k
+^blk\.14\.ffn_gate\.weight$=q3_k
+^blk\.14\.ffn_up\.weight$=q3_k
+^blk\.15\.attn_k\.weight$=q5_k
+^blk\.15\.attn_output\.weight$=q3_k
+^blk\.15\.attn_q\.weight$=q3_k
+^blk\.15\.attn_v\.weight$=q6_k
+^blk\.15\.ffn_down\.weight$=q3_k
+^blk\.15\.ffn_gate\.weight$=q3_k
+^blk\.15\.ffn_up\.weight$=q3_k
+^blk\.16\.attn_k\.weight$=q5_k
+^blk\.16\.attn_output\.weight$=q3_k
+^blk\.16\.attn_q\.weight$=q3_k
+^blk\.16\.attn_v\.weight$=q6_k
+^blk\.16\.ffn_down\.weight$=q3_k
+^blk\.16\.ffn_gate\.weight$=q3_k
+^blk\.16\.ffn_up\.weight$=q3_k
+^blk\.17\.attn_k\.weight$=q5_k
+^blk\.17\.attn_output\.weight$=q3_k
+^blk\.17\.attn_q\.weight$=q3_k
+^blk\.17\.attn_v\.weight$=q6_k
+^blk\.17\.ffn_down\.weight$=q3_k
+^blk\.17\.ffn_gate\.weight$=q3_k
+^blk\.17\.ffn_up\.weight$=q3_k
+^blk\.18\.attn_k\.weight$=q5_k
+^blk\.18\.attn_output\.weight$=q3_k
+^blk\.18\.attn_q\.weight$=q3_k
+^blk\.18\.attn_v\.weight$=q6_k
+^blk\.18\.ffn_down\.weight$=q3_k
+^blk\.18\.ffn_gate\.weight$=q3_k
+^blk\.18\.ffn_up\.weight$=q3_k
+^blk\.19\.attn_k\.weight$=q5_k
+^blk\.19\.attn_output\.weight$=q3_k
+^blk\.19\.attn_q\.weight$=q3_k
+^blk\.19\.attn_v\.weight$=q6_k
+^blk\.19\.ffn_down\.weight$=q3_k
+^blk\.19\.ffn_gate\.weight$=q3_k
+^blk\.19\.ffn_up\.weight$=q3_k
+^blk\.20\.attn_k\.weight$=q5_k
+^blk\.20\.attn_output\.weight$=q3_k
+^blk\.20\.attn_q\.weight$=q3_k
+^blk\.20\.attn_v\.weight$=q6_k
+^blk\.20\.ffn_down\.weight$=q3_k
+^blk\.20\.ffn_gate\.weight$=q3_k
+^blk\.20\.ffn_up\.weight$=q3_k
+^blk\.21\.attn_k\.weight$=q5_k
+^blk\.21\.attn_output\.weight$=q3_k
+^blk\.21\.attn_q\.weight$=q3_k
+^blk\.21\.attn_v\.weight$=q6_k
+^blk\.21\.ffn_down\.weight$=q3_k
+^blk\.21\.ffn_gate\.weight$=q3_k
+^blk\.21\.ffn_up\.weight$=q3_k
+^blk\.22\.attn_k\.weight$=q5_k
+^blk\.22\.attn_output\.weight$=q3_k
+^blk\.22\.attn_q\.weight$=q3_k
+^blk\.22\.attn_v\.weight$=q6_k
+^blk\.22\.ffn_down\.weight$=q3_k
+^blk\.22\.ffn_gate\.weight$=q3_k
+^blk\.22\.ffn_up\.weight$=q3_k
+^blk\.23\.attn_k\.weight$=q5_k
+^blk\.23\.attn_output\.weight$=q3_k
+^blk\.23\.attn_q\.weight$=q3_k
+^blk\.23\.attn_v\.weight$=q6_k
+^blk\.23\.ffn_down\.weight$=q3_k
+^blk\.23\.ffn_gate\.weight$=q3_k
+^blk\.23\.ffn_up\.weight$=q3_k
+^blk\.24\.attn_k\.weight$=q5_k
+^blk\.24\.attn_output\.weight$=q3_k
+^blk\.24\.attn_q\.weight$=q3_k
+^blk\.24\.attn_v\.weight$=q6_k
+^blk\.24\.ffn_down\.weight$=q3_k
+^blk\.24\.ffn_gate\.weight$=q3_k
+^blk\.24\.ffn_up\.weight$=q3_k
+^blk\.25\.attn_k\.weight$=q5_k
+^blk\.25\.attn_output\.weight$=q3_k
+^blk\.25\.attn_q\.weight$=q3_k
+^blk\.25\.attn_v\.weight$=q6_k
+^blk\.25\.ffn_down\.weight$=q3_k
+^blk\.25\.ffn_gate\.weight$=q3_k
+^blk\.25\.ffn_up\.weight$=q3_k
+^blk\.26\.attn_k\.weight$=q5_k
+^blk\.26\.attn_output\.weight$=q3_k
+^blk\.26\.attn_q\.weight$=q3_k
+^blk\.26\.attn_v\.weight$=q6_k
+^blk\.26\.ffn_down\.weight$=q3_k
+^blk\.26\.ffn_gate\.weight$=q3_k
+^blk\.26\.ffn_up\.weight$=q3_k
+^blk\.27\.attn_k\.weight$=q5_k
+^blk\.27\.attn_output\.weight$=q3_k
+^blk\.27\.attn_q\.weight$=q3_k
+^blk\.27\.attn_v\.weight$=q6_k
+^blk\.27\.ffn_down\.weight$=q3_k
+^blk\.27\.ffn_gate\.weight$=q3_k
+^blk\.27\.ffn_up\.weight$=q3_k
+^blk\.28\.attn_k\.weight$=q5_k
+^blk\.28\.attn_output\.weight$=q3_k
+^blk\.28\.attn_q\.weight$=q3_k
+^blk\.28\.attn_v\.weight$=q6_k
+^blk\.28\.ffn_down\.weight$=q3_k
+^blk\.28\.ffn_gate\.weight$=q3_k
+^blk\.28\.ffn_up\.weight$=q3_k
+^blk\.29\.attn_k\.weight$=q5_k
+^blk\.29\.attn_output\.weight$=q3_k
+^blk\.29\.attn_q\.weight$=q3_k
+^blk\.29\.attn_v\.weight$=q6_k
+^blk\.29\.ffn_down\.weight$=q3_k
+^blk\.29\.ffn_gate\.weight$=q3_k
+^blk\.29\.ffn_up\.weight$=q3_k
+^blk\.30\.attn_k\.weight$=q4_k
+^blk\.30\.attn_output\.weight$=q3_k
+^blk\.30\.attn_q\.weight$=q3_k
+^blk\.30\.attn_v\.weight$=q5_k
+^blk\.30\.ffn_down\.weight$=q3_k
+^blk\.30\.ffn_gate\.weight$=q3_k
+^blk\.30\.ffn_up\.weight$=q3_k
+^blk\.31\.attn_k\.weight$=q4_k
+^blk\.31\.attn_output\.weight$=q3_k
+^blk\.31\.attn_q\.weight$=q3_k
+^blk\.31\.attn_v\.weight$=q5_k
+^blk\.31\.ffn_down\.weight$=q3_k
+^blk\.31\.ffn_gate\.weight$=q3_k
+^blk\.31\.ffn_up\.weight$=q3_k
+^blk\.32\.attn_k\.weight$=q4_k
+^blk\.32\.attn_output\.weight$=q3_k
+^blk\.32\.attn_q\.weight$=q3_k
+^blk\.32\.attn_v\.weight$=q5_k
+^blk\.32\.ffn_down\.weight$=q3_k
+^blk\.32\.ffn_gate\.weight$=q3_k
+^blk\.32\.ffn_up\.weight$=q3_k
+^blk\.33\.attn_k\.weight$=q4_k
+^blk\.33\.attn_output\.weight$=q3_k
+^blk\.33\.attn_q\.weight$=q3_k
+^blk\.33\.attn_v\.weight$=q5_k
+^blk\.33\.ffn_down\.weight$=q3_k
+^blk\.33\.ffn_gate\.weight$=q3_k
+^blk\.33\.ffn_up\.weight$=q3_k
+^blk\.34\.attn_k\.weight$=q4_k
+^blk\.34\.attn_output\.weight$=q3_k
+^blk\.34\.attn_q\.weight$=q3_k
+^blk\.34\.attn_v\.weight$=q5_k
+^blk\.34\.ffn_down\.weight$=q3_k
+^blk\.34\.ffn_gate\.weight$=q3_k
+^blk\.34\.ffn_up\.weight$=q3_k
+^blk\.35\.attn_k\.weight$=q4_k
+^blk\.35\.attn_output\.weight$=q3_k
+^blk\.35\.attn_q\.weight$=q3_k
+^blk\.35\.attn_v\.weight$=q5_k
+^blk\.35\.ffn_down\.weight$=q3_k
+^blk\.35\.ffn_gate\.weight$=q3_k
+^blk\.35\.ffn_up\.weight$=q3_k
+^blk\.36\.attn_k\.weight$=q4_k
+^blk\.36\.attn_output\.weight$=q3_k
+^blk\.36\.attn_q\.weight$=q3_k
+^blk\.36\.attn_v\.weight$=q5_k
+^blk\.36\.ffn_down\.weight$=q3_k
+^blk\.36\.ffn_gate\.weight$=q3_k
+^blk\.36\.ffn_up\.weight$=q3_k
+^blk\.37\.attn_k\.weight$=q4_k
+^blk\.37\.attn_output\.weight$=q3_k
+^blk\.37\.attn_q\.weight$=q3_k
+^blk\.37\.attn_v\.weight$=q5_k
+^blk\.37\.ffn_down\.weight$=q3_k
+^blk\.37\.ffn_gate\.weight$=q3_k
+^blk\.37\.ffn_up\.weight$=q3_k
+^blk\.38\.attn_k\.weight$=q4_k
+^blk\.38\.attn_output\.weight$=q3_k
+^blk\.38\.attn_q\.weight$=q3_k
+^blk\.38\.attn_v\.weight$=q5_k
+^blk\.38\.ffn_down\.weight$=q3_k
+^blk\.38\.ffn_gate\.weight$=q3_k
+^blk\.38\.ffn_up\.weight$=q3_k
+^blk\.39\.attn_k\.weight$=q4_k
+^blk\.39\.attn_output\.weight$=q3_k
+^blk\.39\.attn_q\.weight$=q3_k
+^blk\.39\.attn_v\.weight$=q6_k
+^blk\.39\.ffn_down\.weight$=q3_k
+^blk\.39\.ffn_gate\.weight$=q3_k
+^blk\.39\.ffn_up\.weight$=q4_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_L.layout.json

@@ -0,0 +1,88 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q4_K_L",
+  "base_type": "q4_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.50",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "L",
+  "rung_share": 0.5,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0686,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.047,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 9284672280,
+    "map_body+embd": 10008845080,
+    "ratio": 1.078,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.5,
+  "body_bpw": 5.41,
+  "body_bpw_heuristic": 4.972,
+  "total_params": 14765593600,
+  "file_bytes_pred": 10010544920,
+  "file_bpw_pred": 5.424,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q6_k": 22,
+      "q8_0": 9,
+      "q4_k": 9
+    },
+    "attn_q": {
+      "q4_k": 39,
+      "q6_k": 1
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q6_k": 20,
+      "q4_k": 20
+    },
+    "ffn_gate": {
+      "q4_k": 35,
+      "q5_k": 3,
+      "q6_k": 2
+    },
+    "ffn_up": {
+      "q4_k": 20,
+      "q6_k": 19,
+      "q8_0": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q4_K_L",
+  "quantize_ftype": "Q4_K_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_L.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q6_k
+^blk\.0\.attn_q\.weight$=q4_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q6_k
+^blk\.0\.ffn_gate\.weight$=q4_k
+^blk\.0\.ffn_up\.weight$=q6_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q4_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q6_k
+^blk\.1\.ffn_gate\.weight$=q6_k
+^blk\.1\.ffn_up\.weight$=q6_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q4_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q6_k
+^blk\.2\.ffn_gate\.weight$=q5_k
+^blk\.2\.ffn_up\.weight$=q6_k
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q4_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q6_k
+^blk\.3\.ffn_gate\.weight$=q5_k
+^blk\.3\.ffn_up\.weight$=q6_k
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q4_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q6_k
+^blk\.4\.ffn_gate\.weight$=q5_k
+^blk\.4\.ffn_up\.weight$=q6_k
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q4_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q6_k
+^blk\.5\.ffn_gate\.weight$=q4_k
+^blk\.5\.ffn_up\.weight$=q6_k
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q4_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q6_k
+^blk\.6\.ffn_gate\.weight$=q4_k
+^blk\.6\.ffn_up\.weight$=q6_k
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q4_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q6_k
+^blk\.7\.ffn_gate\.weight$=q4_k
+^blk\.7\.ffn_up\.weight$=q6_k
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q4_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q6_k
+^blk\.8\.ffn_gate\.weight$=q4_k
+^blk\.8\.ffn_up\.weight$=q6_k
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q4_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q6_k
+^blk\.9\.ffn_gate\.weight$=q4_k
+^blk\.9\.ffn_up\.weight$=q6_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q6_k
+^blk\.10\.attn_q\.weight$=q6_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q4_k
+^blk\.10\.ffn_gate\.weight$=q4_k
+^blk\.10\.ffn_up\.weight$=q4_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q6_k
+^blk\.11\.attn_q\.weight$=q4_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q4_k
+^blk\.11\.ffn_gate\.weight$=q4_k
+^blk\.11\.ffn_up\.weight$=q4_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q6_k
+^blk\.12\.attn_q\.weight$=q4_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q4_k
+^blk\.12\.ffn_gate\.weight$=q4_k
+^blk\.12\.ffn_up\.weight$=q4_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q6_k
+^blk\.13\.attn_q\.weight$=q4_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q4_k
+^blk\.13\.ffn_gate\.weight$=q4_k
+^blk\.13\.ffn_up\.weight$=q4_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q6_k
+^blk\.14\.attn_q\.weight$=q4_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q4_k
+^blk\.14\.ffn_gate\.weight$=q4_k
+^blk\.14\.ffn_up\.weight$=q4_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q6_k
+^blk\.15\.attn_q\.weight$=q4_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q4_k
+^blk\.15\.ffn_gate\.weight$=q4_k
+^blk\.15\.ffn_up\.weight$=q4_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q6_k
+^blk\.16\.attn_q\.weight$=q4_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q4_k
+^blk\.16\.ffn_gate\.weight$=q4_k
+^blk\.16\.ffn_up\.weight$=q4_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q6_k
+^blk\.17\.attn_q\.weight$=q4_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q4_k
+^blk\.17\.ffn_gate\.weight$=q4_k
+^blk\.17\.ffn_up\.weight$=q4_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q6_k
+^blk\.18\.attn_q\.weight$=q4_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q4_k
+^blk\.18\.ffn_gate\.weight$=q4_k
+^blk\.18\.ffn_up\.weight$=q4_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q6_k
+^blk\.19\.attn_q\.weight$=q4_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q4_k
+^blk\.19\.ffn_gate\.weight$=q4_k
+^blk\.19\.ffn_up\.weight$=q4_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q6_k
+^blk\.20\.attn_q\.weight$=q4_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q4_k
+^blk\.20\.ffn_gate\.weight$=q4_k
+^blk\.20\.ffn_up\.weight$=q4_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q6_k
+^blk\.21\.attn_q\.weight$=q4_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q4_k
+^blk\.21\.ffn_gate\.weight$=q4_k
+^blk\.21\.ffn_up\.weight$=q4_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q6_k
+^blk\.22\.attn_q\.weight$=q4_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q4_k
+^blk\.22\.ffn_gate\.weight$=q4_k
+^blk\.22\.ffn_up\.weight$=q4_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q6_k
+^blk\.23\.attn_q\.weight$=q4_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q4_k
+^blk\.23\.ffn_gate\.weight$=q4_k
+^blk\.23\.ffn_up\.weight$=q4_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q6_k
+^blk\.24\.attn_q\.weight$=q4_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q4_k
+^blk\.24\.ffn_gate\.weight$=q4_k
+^blk\.24\.ffn_up\.weight$=q4_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q6_k
+^blk\.25\.attn_q\.weight$=q4_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q4_k
+^blk\.25\.ffn_gate\.weight$=q4_k
+^blk\.25\.ffn_up\.weight$=q4_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q6_k
+^blk\.26\.attn_q\.weight$=q4_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q4_k
+^blk\.26\.ffn_gate\.weight$=q4_k
+^blk\.26\.ffn_up\.weight$=q4_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q6_k
+^blk\.27\.attn_q\.weight$=q4_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q4_k
+^blk\.27\.ffn_gate\.weight$=q4_k
+^blk\.27\.ffn_up\.weight$=q4_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q6_k
+^blk\.28\.attn_q\.weight$=q4_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q4_k
+^blk\.28\.ffn_gate\.weight$=q4_k
+^blk\.28\.ffn_up\.weight$=q4_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q6_k
+^blk\.29\.attn_q\.weight$=q4_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q4_k
+^blk\.29\.ffn_gate\.weight$=q4_k
+^blk\.29\.ffn_up\.weight$=q4_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q4_k
+^blk\.30\.attn_q\.weight$=q4_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q6_k
+^blk\.30\.ffn_gate\.weight$=q4_k
+^blk\.30\.ffn_up\.weight$=q6_k
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q4_k
+^blk\.31\.attn_q\.weight$=q4_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q6_k
+^blk\.31\.ffn_gate\.weight$=q4_k
+^blk\.31\.ffn_up\.weight$=q6_k
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q4_k
+^blk\.32\.attn_q\.weight$=q4_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q6_k
+^blk\.32\.ffn_gate\.weight$=q4_k
+^blk\.32\.ffn_up\.weight$=q6_k
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q4_k
+^blk\.33\.attn_q\.weight$=q4_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q6_k
+^blk\.33\.ffn_gate\.weight$=q4_k
+^blk\.33\.ffn_up\.weight$=q6_k
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q4_k
+^blk\.34\.attn_q\.weight$=q4_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q6_k
+^blk\.34\.ffn_gate\.weight$=q4_k
+^blk\.34\.ffn_up\.weight$=q6_k
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q4_k
+^blk\.35\.attn_q\.weight$=q4_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q6_k
+^blk\.35\.ffn_gate\.weight$=q4_k
+^blk\.35\.ffn_up\.weight$=q6_k
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q4_k
+^blk\.36\.attn_q\.weight$=q4_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q6_k
+^blk\.36\.ffn_gate\.weight$=q4_k
+^blk\.36\.ffn_up\.weight$=q6_k
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q4_k
+^blk\.37\.attn_q\.weight$=q4_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q6_k
+^blk\.37\.ffn_gate\.weight$=q4_k
+^blk\.37\.ffn_up\.weight$=q6_k
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q4_k
+^blk\.38\.attn_q\.weight$=q4_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q6_k
+^blk\.38\.ffn_gate\.weight$=q4_k
+^blk\.38\.ffn_up\.weight$=q6_k
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q6_k
+^blk\.39\.attn_q\.weight$=q4_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q6_k
+^blk\.39\.ffn_gate\.weight$=q6_k
+^blk\.39\.ffn_up\.weight$=q8_0

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_M.layout.json

@@ -0,0 +1,86 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q4_K_M",
+  "base_type": "q4_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0686,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.047,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 9284672280,
+    "map_body+embd": 9310067480,
+    "ratio": 1.0027,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 4.987,
+  "body_bpw_heuristic": 4.972,
+  "total_params": 14765593600,
+  "file_bytes_pred": 9311767320,
+  "file_bpw_pred": 5.045,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 31,
+      "q6_k": 9
+    },
+    "attn_output": {
+      "q6_k": 20,
+      "q4_k": 11,
+      "q8_0": 9
+    },
+    "attn_q": {
+      "q4_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q4_k": 30,
+      "q5_k": 9,
+      "q6_k": 1
+    },
+    "ffn_gate": {
+      "q4_k": 40
+    },
+    "ffn_up": {
+      "q4_k": 30,
+      "q6_k": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q4_K_M",
+  "quantize_ftype": "Q4_K_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_M.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q4_k
+^blk\.0\.attn_q\.weight$=q4_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q4_k
+^blk\.0\.ffn_gate\.weight$=q4_k
+^blk\.0\.ffn_up\.weight$=q4_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q4_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q5_k
+^blk\.1\.ffn_gate\.weight$=q4_k
+^blk\.1\.ffn_up\.weight$=q6_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q4_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q5_k
+^blk\.2\.ffn_gate\.weight$=q4_k
+^blk\.2\.ffn_up\.weight$=q6_k
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q4_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q5_k
+^blk\.3\.ffn_gate\.weight$=q4_k
+^blk\.3\.ffn_up\.weight$=q6_k
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q4_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q5_k
+^blk\.4\.ffn_gate\.weight$=q4_k
+^blk\.4\.ffn_up\.weight$=q6_k
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q4_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q5_k
+^blk\.5\.ffn_gate\.weight$=q4_k
+^blk\.5\.ffn_up\.weight$=q6_k
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q4_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q5_k
+^blk\.6\.ffn_gate\.weight$=q4_k
+^blk\.6\.ffn_up\.weight$=q6_k
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q4_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q5_k
+^blk\.7\.ffn_gate\.weight$=q4_k
+^blk\.7\.ffn_up\.weight$=q6_k
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q4_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q5_k
+^blk\.8\.ffn_gate\.weight$=q4_k
+^blk\.8\.ffn_up\.weight$=q6_k
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q4_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q5_k
+^blk\.9\.ffn_gate\.weight$=q4_k
+^blk\.9\.ffn_up\.weight$=q6_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q6_k
+^blk\.10\.attn_q\.weight$=q4_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q4_k
+^blk\.10\.ffn_gate\.weight$=q4_k
+^blk\.10\.ffn_up\.weight$=q4_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q6_k
+^blk\.11\.attn_q\.weight$=q4_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q4_k
+^blk\.11\.ffn_gate\.weight$=q4_k
+^blk\.11\.ffn_up\.weight$=q4_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q6_k
+^blk\.12\.attn_q\.weight$=q4_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q4_k
+^blk\.12\.ffn_gate\.weight$=q4_k
+^blk\.12\.ffn_up\.weight$=q4_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q6_k
+^blk\.13\.attn_q\.weight$=q4_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q4_k
+^blk\.13\.ffn_gate\.weight$=q4_k
+^blk\.13\.ffn_up\.weight$=q4_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q6_k
+^blk\.14\.attn_q\.weight$=q4_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q4_k
+^blk\.14\.ffn_gate\.weight$=q4_k
+^blk\.14\.ffn_up\.weight$=q4_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q6_k
+^blk\.15\.attn_q\.weight$=q4_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q4_k
+^blk\.15\.ffn_gate\.weight$=q4_k
+^blk\.15\.ffn_up\.weight$=q4_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q6_k
+^blk\.16\.attn_q\.weight$=q4_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q4_k
+^blk\.16\.ffn_gate\.weight$=q4_k
+^blk\.16\.ffn_up\.weight$=q4_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q6_k
+^blk\.17\.attn_q\.weight$=q4_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q4_k
+^blk\.17\.ffn_gate\.weight$=q4_k
+^blk\.17\.ffn_up\.weight$=q4_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q6_k
+^blk\.18\.attn_q\.weight$=q4_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q4_k
+^blk\.18\.ffn_gate\.weight$=q4_k
+^blk\.18\.ffn_up\.weight$=q4_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q6_k
+^blk\.19\.attn_q\.weight$=q4_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q4_k
+^blk\.19\.ffn_gate\.weight$=q4_k
+^blk\.19\.ffn_up\.weight$=q4_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q6_k
+^blk\.20\.attn_q\.weight$=q4_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q4_k
+^blk\.20\.ffn_gate\.weight$=q4_k
+^blk\.20\.ffn_up\.weight$=q4_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q6_k
+^blk\.21\.attn_q\.weight$=q4_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q4_k
+^blk\.21\.ffn_gate\.weight$=q4_k
+^blk\.21\.ffn_up\.weight$=q4_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q6_k
+^blk\.22\.attn_q\.weight$=q4_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q4_k
+^blk\.22\.ffn_gate\.weight$=q4_k
+^blk\.22\.ffn_up\.weight$=q4_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q6_k
+^blk\.23\.attn_q\.weight$=q4_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q4_k
+^blk\.23\.ffn_gate\.weight$=q4_k
+^blk\.23\.ffn_up\.weight$=q4_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q6_k
+^blk\.24\.attn_q\.weight$=q4_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q4_k
+^blk\.24\.ffn_gate\.weight$=q4_k
+^blk\.24\.ffn_up\.weight$=q4_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q6_k
+^blk\.25\.attn_q\.weight$=q4_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q4_k
+^blk\.25\.ffn_gate\.weight$=q4_k
+^blk\.25\.ffn_up\.weight$=q4_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q6_k
+^blk\.26\.attn_q\.weight$=q4_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q4_k
+^blk\.26\.ffn_gate\.weight$=q4_k
+^blk\.26\.ffn_up\.weight$=q4_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q6_k
+^blk\.27\.attn_q\.weight$=q4_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q4_k
+^blk\.27\.ffn_gate\.weight$=q4_k
+^blk\.27\.ffn_up\.weight$=q4_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q6_k
+^blk\.28\.attn_q\.weight$=q4_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q4_k
+^blk\.28\.ffn_gate\.weight$=q4_k
+^blk\.28\.ffn_up\.weight$=q4_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q6_k
+^blk\.29\.attn_q\.weight$=q4_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q4_k
+^blk\.29\.ffn_gate\.weight$=q4_k
+^blk\.29\.ffn_up\.weight$=q4_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q4_k
+^blk\.30\.attn_q\.weight$=q4_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q4_k
+^blk\.30\.ffn_gate\.weight$=q4_k
+^blk\.30\.ffn_up\.weight$=q4_k
+^blk\.31\.attn_k\.weight$=q6_k
+^blk\.31\.attn_output\.weight$=q4_k
+^blk\.31\.attn_q\.weight$=q4_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q4_k
+^blk\.31\.ffn_gate\.weight$=q4_k
+^blk\.31\.ffn_up\.weight$=q4_k
+^blk\.32\.attn_k\.weight$=q6_k
+^blk\.32\.attn_output\.weight$=q4_k
+^blk\.32\.attn_q\.weight$=q4_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q4_k
+^blk\.32\.ffn_gate\.weight$=q4_k
+^blk\.32\.ffn_up\.weight$=q4_k
+^blk\.33\.attn_k\.weight$=q6_k
+^blk\.33\.attn_output\.weight$=q4_k
+^blk\.33\.attn_q\.weight$=q4_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q4_k
+^blk\.33\.ffn_gate\.weight$=q4_k
+^blk\.33\.ffn_up\.weight$=q4_k
+^blk\.34\.attn_k\.weight$=q6_k
+^blk\.34\.attn_output\.weight$=q4_k
+^blk\.34\.attn_q\.weight$=q4_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q4_k
+^blk\.34\.ffn_gate\.weight$=q4_k
+^blk\.34\.ffn_up\.weight$=q4_k
+^blk\.35\.attn_k\.weight$=q6_k
+^blk\.35\.attn_output\.weight$=q4_k
+^blk\.35\.attn_q\.weight$=q4_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q4_k
+^blk\.35\.ffn_gate\.weight$=q4_k
+^blk\.35\.ffn_up\.weight$=q4_k
+^blk\.36\.attn_k\.weight$=q6_k
+^blk\.36\.attn_output\.weight$=q4_k
+^blk\.36\.attn_q\.weight$=q4_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q4_k
+^blk\.36\.ffn_gate\.weight$=q4_k
+^blk\.36\.ffn_up\.weight$=q4_k
+^blk\.37\.attn_k\.weight$=q6_k
+^blk\.37\.attn_output\.weight$=q4_k
+^blk\.37\.attn_q\.weight$=q4_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q4_k
+^blk\.37\.ffn_gate\.weight$=q4_k
+^blk\.37\.ffn_up\.weight$=q4_k
+^blk\.38\.attn_k\.weight$=q6_k
+^blk\.38\.attn_output\.weight$=q4_k
+^blk\.38\.attn_q\.weight$=q4_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q4_k
+^blk\.38\.ffn_gate\.weight$=q4_k
+^blk\.38\.ffn_up\.weight$=q4_k
+^blk\.39\.attn_k\.weight$=q6_k
+^blk\.39\.attn_output\.weight$=q4_k
+^blk\.39\.attn_q\.weight$=q4_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q6_k
+^blk\.39\.ffn_gate\.weight$=q4_k
+^blk\.39\.ffn_up\.weight$=q6_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_S.layout.json

@@ -0,0 +1,84 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q4_K_S",
+  "base_type": "q4_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0731,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0501,
+      "type": "q4_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q4_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 8709593880,
+    "map_body+embd": 8849062680,
+    "ratio": 1.016,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.9,
+  "body_bpw": 4.708,
+  "body_bpw_heuristic": 4.624,
+  "total_params": 14765593600,
+  "file_bytes_pred": 8850762520,
+  "file_bpw_pred": 4.795,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 29,
+      "q6_k": 11
+    },
+    "attn_output": {
+      "q4_k": 30,
+      "q8_0": 9,
+      "q6_k": 1
+    },
+    "attn_q": {
+      "q4_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q4_k": 40
+    },
+    "ffn_gate": {
+      "q4_k": 40
+    },
+    "ffn_up": {
+      "q4_k": 39,
+      "q6_k": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q4_K_S",
+  "quantize_ftype": "Q4_K_S",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q4_K_S.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q4_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q4_k
+^blk\.0\.attn_q\.weight$=q4_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q4_k
+^blk\.0\.ffn_gate\.weight$=q4_k
+^blk\.0\.ffn_up\.weight$=q4_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q4_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q4_k
+^blk\.1\.ffn_gate\.weight$=q4_k
+^blk\.1\.ffn_up\.weight$=q4_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q4_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q4_k
+^blk\.2\.ffn_gate\.weight$=q4_k
+^blk\.2\.ffn_up\.weight$=q4_k
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q4_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q4_k
+^blk\.3\.ffn_gate\.weight$=q4_k
+^blk\.3\.ffn_up\.weight$=q4_k
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q4_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q4_k
+^blk\.4\.ffn_gate\.weight$=q4_k
+^blk\.4\.ffn_up\.weight$=q4_k
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q4_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q4_k
+^blk\.5\.ffn_gate\.weight$=q4_k
+^blk\.5\.ffn_up\.weight$=q4_k
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q4_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q4_k
+^blk\.6\.ffn_gate\.weight$=q4_k
+^blk\.6\.ffn_up\.weight$=q4_k
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q4_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q4_k
+^blk\.7\.ffn_gate\.weight$=q4_k
+^blk\.7\.ffn_up\.weight$=q4_k
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q4_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q4_k
+^blk\.8\.ffn_gate\.weight$=q4_k
+^blk\.8\.ffn_up\.weight$=q4_k
+^blk\.9\.attn_k\.weight$=q6_k
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q4_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q4_k
+^blk\.9\.ffn_gate\.weight$=q4_k
+^blk\.9\.ffn_up\.weight$=q4_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q6_k
+^blk\.10\.attn_q\.weight$=q4_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q4_k
+^blk\.10\.ffn_gate\.weight$=q4_k
+^blk\.10\.ffn_up\.weight$=q4_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q4_k
+^blk\.11\.attn_q\.weight$=q4_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q4_k
+^blk\.11\.ffn_gate\.weight$=q4_k
+^blk\.11\.ffn_up\.weight$=q4_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q4_k
+^blk\.12\.attn_q\.weight$=q4_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q4_k
+^blk\.12\.ffn_gate\.weight$=q4_k
+^blk\.12\.ffn_up\.weight$=q4_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q4_k
+^blk\.13\.attn_q\.weight$=q4_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q4_k
+^blk\.13\.ffn_gate\.weight$=q4_k
+^blk\.13\.ffn_up\.weight$=q4_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q4_k
+^blk\.14\.attn_q\.weight$=q4_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q4_k
+^blk\.14\.ffn_gate\.weight$=q4_k
+^blk\.14\.ffn_up\.weight$=q4_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q4_k
+^blk\.15\.attn_q\.weight$=q4_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q4_k
+^blk\.15\.ffn_gate\.weight$=q4_k
+^blk\.15\.ffn_up\.weight$=q4_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q4_k
+^blk\.16\.attn_q\.weight$=q4_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q4_k
+^blk\.16\.ffn_gate\.weight$=q4_k
+^blk\.16\.ffn_up\.weight$=q4_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q4_k
+^blk\.17\.attn_q\.weight$=q4_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q4_k
+^blk\.17\.ffn_gate\.weight$=q4_k
+^blk\.17\.ffn_up\.weight$=q4_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q4_k
+^blk\.18\.attn_q\.weight$=q4_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q4_k
+^blk\.18\.ffn_gate\.weight$=q4_k
+^blk\.18\.ffn_up\.weight$=q4_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q4_k
+^blk\.19\.attn_q\.weight$=q4_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q4_k
+^blk\.19\.ffn_gate\.weight$=q4_k
+^blk\.19\.ffn_up\.weight$=q4_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q4_k
+^blk\.20\.attn_q\.weight$=q4_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q4_k
+^blk\.20\.ffn_gate\.weight$=q4_k
+^blk\.20\.ffn_up\.weight$=q4_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q4_k
+^blk\.21\.attn_q\.weight$=q4_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q4_k
+^blk\.21\.ffn_gate\.weight$=q4_k
+^blk\.21\.ffn_up\.weight$=q4_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q4_k
+^blk\.22\.attn_q\.weight$=q4_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q4_k
+^blk\.22\.ffn_gate\.weight$=q4_k
+^blk\.22\.ffn_up\.weight$=q4_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q4_k
+^blk\.23\.attn_q\.weight$=q4_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q4_k
+^blk\.23\.ffn_gate\.weight$=q4_k
+^blk\.23\.ffn_up\.weight$=q4_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q4_k
+^blk\.24\.attn_q\.weight$=q4_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q4_k
+^blk\.24\.ffn_gate\.weight$=q4_k
+^blk\.24\.ffn_up\.weight$=q4_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q4_k
+^blk\.25\.attn_q\.weight$=q4_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q4_k
+^blk\.25\.ffn_gate\.weight$=q4_k
+^blk\.25\.ffn_up\.weight$=q4_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q4_k
+^blk\.26\.attn_q\.weight$=q4_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q4_k
+^blk\.26\.ffn_gate\.weight$=q4_k
+^blk\.26\.ffn_up\.weight$=q4_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q4_k
+^blk\.27\.attn_q\.weight$=q4_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q4_k
+^blk\.27\.ffn_gate\.weight$=q4_k
+^blk\.27\.ffn_up\.weight$=q4_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q4_k
+^blk\.28\.attn_q\.weight$=q4_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q4_k
+^blk\.28\.ffn_gate\.weight$=q4_k
+^blk\.28\.ffn_up\.weight$=q4_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q4_k
+^blk\.29\.attn_q\.weight$=q4_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q4_k
+^blk\.29\.ffn_gate\.weight$=q4_k
+^blk\.29\.ffn_up\.weight$=q4_k
+^blk\.30\.attn_k\.weight$=q6_k
+^blk\.30\.attn_output\.weight$=q4_k
+^blk\.30\.attn_q\.weight$=q4_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q4_k
+^blk\.30\.ffn_gate\.weight$=q4_k
+^blk\.30\.ffn_up\.weight$=q4_k
+^blk\.31\.attn_k\.weight$=q6_k
+^blk\.31\.attn_output\.weight$=q4_k
+^blk\.31\.attn_q\.weight$=q4_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q4_k
+^blk\.31\.ffn_gate\.weight$=q4_k
+^blk\.31\.ffn_up\.weight$=q4_k
+^blk\.32\.attn_k\.weight$=q6_k
+^blk\.32\.attn_output\.weight$=q4_k
+^blk\.32\.attn_q\.weight$=q4_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q4_k
+^blk\.32\.ffn_gate\.weight$=q4_k
+^blk\.32\.ffn_up\.weight$=q4_k
+^blk\.33\.attn_k\.weight$=q6_k
+^blk\.33\.attn_output\.weight$=q4_k
+^blk\.33\.attn_q\.weight$=q4_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q4_k
+^blk\.33\.ffn_gate\.weight$=q4_k
+^blk\.33\.ffn_up\.weight$=q4_k
+^blk\.34\.attn_k\.weight$=q6_k
+^blk\.34\.attn_output\.weight$=q4_k
+^blk\.34\.attn_q\.weight$=q4_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q4_k
+^blk\.34\.ffn_gate\.weight$=q4_k
+^blk\.34\.ffn_up\.weight$=q4_k
+^blk\.35\.attn_k\.weight$=q6_k
+^blk\.35\.attn_output\.weight$=q4_k
+^blk\.35\.attn_q\.weight$=q4_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q4_k
+^blk\.35\.ffn_gate\.weight$=q4_k
+^blk\.35\.ffn_up\.weight$=q4_k
+^blk\.36\.attn_k\.weight$=q6_k
+^blk\.36\.attn_output\.weight$=q4_k
+^blk\.36\.attn_q\.weight$=q4_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q4_k
+^blk\.36\.ffn_gate\.weight$=q4_k
+^blk\.36\.ffn_up\.weight$=q4_k
+^blk\.37\.attn_k\.weight$=q6_k
+^blk\.37\.attn_output\.weight$=q4_k
+^blk\.37\.attn_q\.weight$=q4_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q4_k
+^blk\.37\.ffn_gate\.weight$=q4_k
+^blk\.37\.ffn_up\.weight$=q4_k
+^blk\.38\.attn_k\.weight$=q6_k
+^blk\.38\.attn_output\.weight$=q4_k
+^blk\.38\.attn_q\.weight$=q4_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q4_k
+^blk\.38\.ffn_gate\.weight$=q4_k
+^blk\.38\.ffn_up\.weight$=q4_k
+^blk\.39\.attn_k\.weight$=q6_k
+^blk\.39\.attn_output\.weight$=q4_k
+^blk\.39\.attn_q\.weight$=q4_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q4_k
+^blk\.39\.ffn_gate\.weight$=q4_k
+^blk\.39\.ffn_up\.weight$=q6_k

layouts/togethercomputer_HybridAgent-14B-Agentic-Q5_K_M.layout.json

@@ -0,0 +1,85 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q5_K_M",
+  "base_type": "q5_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0593,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0497,
+      "type": "q5_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q5_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 10741613720,
+    "map_body+embd": 11184063640,
+    "ratio": 1.0412,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 6.063,
+  "body_bpw_heuristic": 5.795,
+  "total_params": 14765593600,
+  "file_bytes_pred": 11185763480,
+  "file_bpw_pred": 6.06,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q8_0": 30,
+      "q5_k": 9,
+      "q6_k": 1
+    },
+    "attn_q": {
+      "q5_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q5_k": 30,
+      "q6_k": 9,
+      "q8_0": 1
+    },
+    "ffn_gate": {
+      "q5_k": 40
+    },
+    "ffn_up": {
+      "q5_k": 30,
+      "q8_0": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q5_K_M",
+  "quantize_ftype": "Q5_K_M",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q5_K_M.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q5_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q8_0
+^blk\.0\.attn_q\.weight$=q5_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q5_k
+^blk\.0\.ffn_gate\.weight$=q5_k
+^blk\.0\.ffn_up\.weight$=q5_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q5_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q6_k
+^blk\.1\.ffn_gate\.weight$=q5_k
+^blk\.1\.ffn_up\.weight$=q8_0
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q5_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q6_k
+^blk\.2\.ffn_gate\.weight$=q5_k
+^blk\.2\.ffn_up\.weight$=q8_0
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q5_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q6_k
+^blk\.3\.ffn_gate\.weight$=q5_k
+^blk\.3\.ffn_up\.weight$=q8_0
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q5_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q6_k
+^blk\.4\.ffn_gate\.weight$=q5_k
+^blk\.4\.ffn_up\.weight$=q8_0
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q5_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q6_k
+^blk\.5\.ffn_gate\.weight$=q5_k
+^blk\.5\.ffn_up\.weight$=q8_0
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q5_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q6_k
+^blk\.6\.ffn_gate\.weight$=q5_k
+^blk\.6\.ffn_up\.weight$=q8_0
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q5_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q6_k
+^blk\.7\.ffn_gate\.weight$=q5_k
+^blk\.7\.ffn_up\.weight$=q8_0
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q5_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q6_k
+^blk\.8\.ffn_gate\.weight$=q5_k
+^blk\.8\.ffn_up\.weight$=q8_0
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q5_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q6_k
+^blk\.9\.ffn_gate\.weight$=q5_k
+^blk\.9\.ffn_up\.weight$=q8_0
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q8_0
+^blk\.10\.attn_q\.weight$=q5_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q5_k
+^blk\.10\.ffn_gate\.weight$=q5_k
+^blk\.10\.ffn_up\.weight$=q5_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q8_0
+^blk\.11\.attn_q\.weight$=q5_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q5_k
+^blk\.11\.ffn_gate\.weight$=q5_k
+^blk\.11\.ffn_up\.weight$=q5_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q8_0
+^blk\.12\.attn_q\.weight$=q5_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q5_k
+^blk\.12\.ffn_gate\.weight$=q5_k
+^blk\.12\.ffn_up\.weight$=q5_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q8_0
+^blk\.13\.attn_q\.weight$=q5_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q5_k
+^blk\.13\.ffn_gate\.weight$=q5_k
+^blk\.13\.ffn_up\.weight$=q5_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q8_0
+^blk\.14\.attn_q\.weight$=q5_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q5_k
+^blk\.14\.ffn_gate\.weight$=q5_k
+^blk\.14\.ffn_up\.weight$=q5_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q8_0
+^blk\.15\.attn_q\.weight$=q5_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q5_k
+^blk\.15\.ffn_gate\.weight$=q5_k
+^blk\.15\.ffn_up\.weight$=q5_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q8_0
+^blk\.16\.attn_q\.weight$=q5_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q5_k
+^blk\.16\.ffn_gate\.weight$=q5_k
+^blk\.16\.ffn_up\.weight$=q5_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q8_0
+^blk\.17\.attn_q\.weight$=q5_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q5_k
+^blk\.17\.ffn_gate\.weight$=q5_k
+^blk\.17\.ffn_up\.weight$=q5_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q8_0
+^blk\.18\.attn_q\.weight$=q5_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q5_k
+^blk\.18\.ffn_gate\.weight$=q5_k
+^blk\.18\.ffn_up\.weight$=q5_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q8_0
+^blk\.19\.attn_q\.weight$=q5_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q5_k
+^blk\.19\.ffn_gate\.weight$=q5_k
+^blk\.19\.ffn_up\.weight$=q5_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q8_0
+^blk\.20\.attn_q\.weight$=q5_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q5_k
+^blk\.20\.ffn_gate\.weight$=q5_k
+^blk\.20\.ffn_up\.weight$=q5_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q8_0
+^blk\.21\.attn_q\.weight$=q5_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q5_k
+^blk\.21\.ffn_gate\.weight$=q5_k
+^blk\.21\.ffn_up\.weight$=q5_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q8_0
+^blk\.22\.attn_q\.weight$=q5_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q5_k
+^blk\.22\.ffn_gate\.weight$=q5_k
+^blk\.22\.ffn_up\.weight$=q5_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q8_0
+^blk\.23\.attn_q\.weight$=q5_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q5_k
+^blk\.23\.ffn_gate\.weight$=q5_k
+^blk\.23\.ffn_up\.weight$=q5_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q8_0
+^blk\.24\.attn_q\.weight$=q5_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q5_k
+^blk\.24\.ffn_gate\.weight$=q5_k
+^blk\.24\.ffn_up\.weight$=q5_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q8_0
+^blk\.25\.attn_q\.weight$=q5_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q5_k
+^blk\.25\.ffn_gate\.weight$=q5_k
+^blk\.25\.ffn_up\.weight$=q5_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q8_0
+^blk\.26\.attn_q\.weight$=q5_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q5_k
+^blk\.26\.ffn_gate\.weight$=q5_k
+^blk\.26\.ffn_up\.weight$=q5_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q8_0
+^blk\.27\.attn_q\.weight$=q5_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q5_k
+^blk\.27\.ffn_gate\.weight$=q5_k
+^blk\.27\.ffn_up\.weight$=q5_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q8_0
+^blk\.28\.attn_q\.weight$=q5_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q5_k
+^blk\.28\.ffn_gate\.weight$=q5_k
+^blk\.28\.ffn_up\.weight$=q5_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q8_0
+^blk\.29\.attn_q\.weight$=q5_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q5_k
+^blk\.29\.ffn_gate\.weight$=q5_k
+^blk\.29\.ffn_up\.weight$=q5_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q5_k
+^blk\.30\.attn_q\.weight$=q5_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q5_k
+^blk\.30\.ffn_gate\.weight$=q5_k
+^blk\.30\.ffn_up\.weight$=q5_k
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q5_k
+^blk\.31\.attn_q\.weight$=q5_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q5_k
+^blk\.31\.ffn_gate\.weight$=q5_k
+^blk\.31\.ffn_up\.weight$=q5_k
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q5_k
+^blk\.32\.attn_q\.weight$=q5_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q5_k
+^blk\.32\.ffn_gate\.weight$=q5_k
+^blk\.32\.ffn_up\.weight$=q5_k
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q5_k
+^blk\.33\.attn_q\.weight$=q5_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q5_k
+^blk\.33\.ffn_gate\.weight$=q5_k
+^blk\.33\.ffn_up\.weight$=q5_k
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q5_k
+^blk\.34\.attn_q\.weight$=q5_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q5_k
+^blk\.34\.ffn_gate\.weight$=q5_k
+^blk\.34\.ffn_up\.weight$=q5_k
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q5_k
+^blk\.35\.attn_q\.weight$=q5_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q5_k
+^blk\.35\.ffn_gate\.weight$=q5_k
+^blk\.35\.ffn_up\.weight$=q5_k
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q5_k
+^blk\.36\.attn_q\.weight$=q5_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q5_k
+^blk\.36\.ffn_gate\.weight$=q5_k
+^blk\.36\.ffn_up\.weight$=q5_k
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q5_k
+^blk\.37\.attn_q\.weight$=q5_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q5_k
+^blk\.37\.ffn_gate\.weight$=q5_k
+^blk\.37\.ffn_up\.weight$=q5_k
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q5_k
+^blk\.38\.attn_q\.weight$=q5_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q5_k
+^blk\.38\.ffn_gate\.weight$=q5_k
+^blk\.38\.ffn_up\.weight$=q5_k
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q6_k
+^blk\.39\.attn_q\.weight$=q5_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q8_0
+^blk\.39\.ffn_gate\.weight$=q5_k
+^blk\.39\.ffn_up\.weight$=q8_0

layouts/togethercomputer_HybridAgent-14B-Agentic-Q5_K_S.layout.json

@@ -0,0 +1,83 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q5_K_S",
+  "base_type": "q5_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0617,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0517,
+      "type": "q5_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q5_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 10323821720,
+    "map_body+embd": 10573309080,
+    "ratio": 1.0242,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.9,
+  "body_bpw": 5.693,
+  "body_bpw_heuristic": 5.542,
+  "total_params": 14765593600,
+  "file_bytes_pred": 10575008920,
+  "file_bpw_pred": 5.73,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q5_k": 25,
+      "q8_0": 12,
+      "q6_k": 3
+    },
+    "attn_q": {
+      "q5_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q5_k": 40
+    },
+    "ffn_gate": {
+      "q5_k": 40
+    },
+    "ffn_up": {
+      "q5_k": 39,
+      "q8_0": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q5_K_S",
+  "quantize_ftype": "Q5_K_S",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q5_K_S.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q5_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q5_k
+^blk\.0\.attn_q\.weight$=q5_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q5_k
+^blk\.0\.ffn_gate\.weight$=q5_k
+^blk\.0\.ffn_up\.weight$=q5_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q5_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q5_k
+^blk\.1\.ffn_gate\.weight$=q5_k
+^blk\.1\.ffn_up\.weight$=q5_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q5_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q5_k
+^blk\.2\.ffn_gate\.weight$=q5_k
+^blk\.2\.ffn_up\.weight$=q5_k
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q5_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q5_k
+^blk\.3\.ffn_gate\.weight$=q5_k
+^blk\.3\.ffn_up\.weight$=q5_k
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q5_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q5_k
+^blk\.4\.ffn_gate\.weight$=q5_k
+^blk\.4\.ffn_up\.weight$=q5_k
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q5_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q5_k
+^blk\.5\.ffn_gate\.weight$=q5_k
+^blk\.5\.ffn_up\.weight$=q5_k
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q5_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q5_k
+^blk\.6\.ffn_gate\.weight$=q5_k
+^blk\.6\.ffn_up\.weight$=q5_k
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q5_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q5_k
+^blk\.7\.ffn_gate\.weight$=q5_k
+^blk\.7\.ffn_up\.weight$=q5_k
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q5_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q5_k
+^blk\.8\.ffn_gate\.weight$=q5_k
+^blk\.8\.ffn_up\.weight$=q5_k
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q5_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q5_k
+^blk\.9\.ffn_gate\.weight$=q5_k
+^blk\.9\.ffn_up\.weight$=q5_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q8_0
+^blk\.10\.attn_q\.weight$=q5_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q5_k
+^blk\.10\.ffn_gate\.weight$=q5_k
+^blk\.10\.ffn_up\.weight$=q5_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q8_0
+^blk\.11\.attn_q\.weight$=q5_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q5_k
+^blk\.11\.ffn_gate\.weight$=q5_k
+^blk\.11\.ffn_up\.weight$=q5_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q8_0
+^blk\.12\.attn_q\.weight$=q5_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q5_k
+^blk\.12\.ffn_gate\.weight$=q5_k
+^blk\.12\.ffn_up\.weight$=q5_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q6_k
+^blk\.13\.attn_q\.weight$=q5_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q5_k
+^blk\.13\.ffn_gate\.weight$=q5_k
+^blk\.13\.ffn_up\.weight$=q5_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q6_k
+^blk\.14\.attn_q\.weight$=q5_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q5_k
+^blk\.14\.ffn_gate\.weight$=q5_k
+^blk\.14\.ffn_up\.weight$=q5_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q6_k
+^blk\.15\.attn_q\.weight$=q5_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q5_k
+^blk\.15\.ffn_gate\.weight$=q5_k
+^blk\.15\.ffn_up\.weight$=q5_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q5_k
+^blk\.16\.attn_q\.weight$=q5_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q5_k
+^blk\.16\.ffn_gate\.weight$=q5_k
+^blk\.16\.ffn_up\.weight$=q5_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q5_k
+^blk\.17\.attn_q\.weight$=q5_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q5_k
+^blk\.17\.ffn_gate\.weight$=q5_k
+^blk\.17\.ffn_up\.weight$=q5_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q5_k
+^blk\.18\.attn_q\.weight$=q5_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q5_k
+^blk\.18\.ffn_gate\.weight$=q5_k
+^blk\.18\.ffn_up\.weight$=q5_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q5_k
+^blk\.19\.attn_q\.weight$=q5_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q5_k
+^blk\.19\.ffn_gate\.weight$=q5_k
+^blk\.19\.ffn_up\.weight$=q5_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q5_k
+^blk\.20\.attn_q\.weight$=q5_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q5_k
+^blk\.20\.ffn_gate\.weight$=q5_k
+^blk\.20\.ffn_up\.weight$=q5_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q5_k
+^blk\.21\.attn_q\.weight$=q5_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q5_k
+^blk\.21\.ffn_gate\.weight$=q5_k
+^blk\.21\.ffn_up\.weight$=q5_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q5_k
+^blk\.22\.attn_q\.weight$=q5_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q5_k
+^blk\.22\.ffn_gate\.weight$=q5_k
+^blk\.22\.ffn_up\.weight$=q5_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q5_k
+^blk\.23\.attn_q\.weight$=q5_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q5_k
+^blk\.23\.ffn_gate\.weight$=q5_k
+^blk\.23\.ffn_up\.weight$=q5_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q5_k
+^blk\.24\.attn_q\.weight$=q5_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q5_k
+^blk\.24\.ffn_gate\.weight$=q5_k
+^blk\.24\.ffn_up\.weight$=q5_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q5_k
+^blk\.25\.attn_q\.weight$=q5_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q5_k
+^blk\.25\.ffn_gate\.weight$=q5_k
+^blk\.25\.ffn_up\.weight$=q5_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q5_k
+^blk\.26\.attn_q\.weight$=q5_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q5_k
+^blk\.26\.ffn_gate\.weight$=q5_k
+^blk\.26\.ffn_up\.weight$=q5_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q5_k
+^blk\.27\.attn_q\.weight$=q5_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q5_k
+^blk\.27\.ffn_gate\.weight$=q5_k
+^blk\.27\.ffn_up\.weight$=q5_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q5_k
+^blk\.28\.attn_q\.weight$=q5_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q5_k
+^blk\.28\.ffn_gate\.weight$=q5_k
+^blk\.28\.ffn_up\.weight$=q5_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q5_k
+^blk\.29\.attn_q\.weight$=q5_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q5_k
+^blk\.29\.ffn_gate\.weight$=q5_k
+^blk\.29\.ffn_up\.weight$=q5_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q5_k
+^blk\.30\.attn_q\.weight$=q5_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q5_k
+^blk\.30\.ffn_gate\.weight$=q5_k
+^blk\.30\.ffn_up\.weight$=q5_k
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q5_k
+^blk\.31\.attn_q\.weight$=q5_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q5_k
+^blk\.31\.ffn_gate\.weight$=q5_k
+^blk\.31\.ffn_up\.weight$=q5_k
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q5_k
+^blk\.32\.attn_q\.weight$=q5_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q5_k
+^blk\.32\.ffn_gate\.weight$=q5_k
+^blk\.32\.ffn_up\.weight$=q5_k
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q5_k
+^blk\.33\.attn_q\.weight$=q5_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q5_k
+^blk\.33\.ffn_gate\.weight$=q5_k
+^blk\.33\.ffn_up\.weight$=q5_k
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q5_k
+^blk\.34\.attn_q\.weight$=q5_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q5_k
+^blk\.34\.ffn_gate\.weight$=q5_k
+^blk\.34\.ffn_up\.weight$=q5_k
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q5_k
+^blk\.35\.attn_q\.weight$=q5_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q5_k
+^blk\.35\.ffn_gate\.weight$=q5_k
+^blk\.35\.ffn_up\.weight$=q5_k
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q5_k
+^blk\.36\.attn_q\.weight$=q5_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q5_k
+^blk\.36\.ffn_gate\.weight$=q5_k
+^blk\.36\.ffn_up\.weight$=q5_k
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q5_k
+^blk\.37\.attn_q\.weight$=q5_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q5_k
+^blk\.37\.ffn_gate\.weight$=q5_k
+^blk\.37\.ffn_up\.weight$=q5_k
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q5_k
+^blk\.38\.attn_q\.weight$=q5_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q5_k
+^blk\.38\.ffn_gate\.weight$=q5_k
+^blk\.38\.ffn_up\.weight$=q5_k
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q5_k
+^blk\.39\.attn_q\.weight$=q5_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q5_k
+^blk\.39\.ffn_gate\.weight$=q5_k
+^blk\.39\.ffn_up\.weight$=q8_0

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K.layout.json

@@ -0,0 +1,84 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q6_K",
+  "base_type": "q6_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.70",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "M",
+  "rung_share": 0.7,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q6_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 12569166000,
+    "map_body+embd": 12906922160,
+    "ratio": 1.0269,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.7,
+  "body_bpw": 7.044,
+  "body_bpw_heuristic": 6.839,
+  "total_params": 14765593600,
+  "file_bytes_pred": 12908622000,
+  "file_bpw_pred": 6.994,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q8_0": 31,
+      "q6_k": 9
+    },
+    "attn_q": {
+      "q6_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q6_k": 28,
+      "q8_0": 12
+    },
+    "ffn_gate": {
+      "q6_k": 39,
+      "q8_0": 1
+    },
+    "ffn_up": {
+      "q6_k": 30,
+      "q8_0": 10
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q6_K",
+  "quantize_ftype": "Q6_K",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q6_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q8_0
+^blk\.0\.attn_q\.weight$=q6_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q8_0
+^blk\.0\.ffn_gate\.weight$=q6_k
+^blk\.0\.ffn_up\.weight$=q6_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q6_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q8_0
+^blk\.1\.ffn_gate\.weight$=q6_k
+^blk\.1\.ffn_up\.weight$=q8_0
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q6_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q8_0
+^blk\.2\.ffn_gate\.weight$=q6_k
+^blk\.2\.ffn_up\.weight$=q8_0
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q6_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q8_0
+^blk\.3\.ffn_gate\.weight$=q6_k
+^blk\.3\.ffn_up\.weight$=q8_0
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q6_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q8_0
+^blk\.4\.ffn_gate\.weight$=q6_k
+^blk\.4\.ffn_up\.weight$=q8_0
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q6_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q8_0
+^blk\.5\.ffn_gate\.weight$=q6_k
+^blk\.5\.ffn_up\.weight$=q8_0
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q6_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q8_0
+^blk\.6\.ffn_gate\.weight$=q6_k
+^blk\.6\.ffn_up\.weight$=q8_0
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q6_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q8_0
+^blk\.7\.ffn_gate\.weight$=q6_k
+^blk\.7\.ffn_up\.weight$=q8_0
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q6_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q8_0
+^blk\.8\.ffn_gate\.weight$=q6_k
+^blk\.8\.ffn_up\.weight$=q8_0
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q6_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q8_0
+^blk\.9\.ffn_gate\.weight$=q6_k
+^blk\.9\.ffn_up\.weight$=q8_0
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q8_0
+^blk\.10\.attn_q\.weight$=q6_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q6_k
+^blk\.10\.ffn_gate\.weight$=q6_k
+^blk\.10\.ffn_up\.weight$=q6_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q8_0
+^blk\.11\.attn_q\.weight$=q6_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q6_k
+^blk\.11\.ffn_gate\.weight$=q6_k
+^blk\.11\.ffn_up\.weight$=q6_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q8_0
+^blk\.12\.attn_q\.weight$=q6_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q6_k
+^blk\.12\.ffn_gate\.weight$=q6_k
+^blk\.12\.ffn_up\.weight$=q6_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q8_0
+^blk\.13\.attn_q\.weight$=q6_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q6_k
+^blk\.13\.ffn_gate\.weight$=q6_k
+^blk\.13\.ffn_up\.weight$=q6_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q8_0
+^blk\.14\.attn_q\.weight$=q6_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q6_k
+^blk\.14\.ffn_gate\.weight$=q6_k
+^blk\.14\.ffn_up\.weight$=q6_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q8_0
+^blk\.15\.attn_q\.weight$=q6_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q6_k
+^blk\.15\.ffn_gate\.weight$=q6_k
+^blk\.15\.ffn_up\.weight$=q6_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q8_0
+^blk\.16\.attn_q\.weight$=q6_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q6_k
+^blk\.16\.ffn_gate\.weight$=q6_k
+^blk\.16\.ffn_up\.weight$=q6_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q8_0
+^blk\.17\.attn_q\.weight$=q6_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q6_k
+^blk\.17\.ffn_gate\.weight$=q6_k
+^blk\.17\.ffn_up\.weight$=q6_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q8_0
+^blk\.18\.attn_q\.weight$=q6_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q6_k
+^blk\.18\.ffn_gate\.weight$=q6_k
+^blk\.18\.ffn_up\.weight$=q6_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q8_0
+^blk\.19\.attn_q\.weight$=q6_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q6_k
+^blk\.19\.ffn_gate\.weight$=q6_k
+^blk\.19\.ffn_up\.weight$=q6_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q8_0
+^blk\.20\.attn_q\.weight$=q6_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q6_k
+^blk\.20\.ffn_gate\.weight$=q6_k
+^blk\.20\.ffn_up\.weight$=q6_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q8_0
+^blk\.21\.attn_q\.weight$=q6_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q6_k
+^blk\.21\.ffn_gate\.weight$=q6_k
+^blk\.21\.ffn_up\.weight$=q6_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q8_0
+^blk\.22\.attn_q\.weight$=q6_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q6_k
+^blk\.22\.ffn_gate\.weight$=q6_k
+^blk\.22\.ffn_up\.weight$=q6_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q8_0
+^blk\.23\.attn_q\.weight$=q6_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q6_k
+^blk\.23\.ffn_gate\.weight$=q6_k
+^blk\.23\.ffn_up\.weight$=q6_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q8_0
+^blk\.24\.attn_q\.weight$=q6_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q6_k
+^blk\.24\.ffn_gate\.weight$=q6_k
+^blk\.24\.ffn_up\.weight$=q6_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q8_0
+^blk\.25\.attn_q\.weight$=q6_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q6_k
+^blk\.25\.ffn_gate\.weight$=q6_k
+^blk\.25\.ffn_up\.weight$=q6_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q8_0
+^blk\.26\.attn_q\.weight$=q6_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q6_k
+^blk\.26\.ffn_gate\.weight$=q6_k
+^blk\.26\.ffn_up\.weight$=q6_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q8_0
+^blk\.27\.attn_q\.weight$=q6_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q6_k
+^blk\.27\.ffn_gate\.weight$=q6_k
+^blk\.27\.ffn_up\.weight$=q6_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q8_0
+^blk\.28\.attn_q\.weight$=q6_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q6_k
+^blk\.28\.ffn_gate\.weight$=q6_k
+^blk\.28\.ffn_up\.weight$=q6_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q8_0
+^blk\.29\.attn_q\.weight$=q6_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q6_k
+^blk\.29\.ffn_gate\.weight$=q6_k
+^blk\.29\.ffn_up\.weight$=q6_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q6_k
+^blk\.30\.attn_q\.weight$=q6_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q8_0
+^blk\.30\.ffn_gate\.weight$=q6_k
+^blk\.30\.ffn_up\.weight$=q6_k
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q6_k
+^blk\.31\.attn_q\.weight$=q6_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q6_k
+^blk\.31\.ffn_gate\.weight$=q6_k
+^blk\.31\.ffn_up\.weight$=q6_k
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q6_k
+^blk\.32\.attn_q\.weight$=q6_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q6_k
+^blk\.32\.ffn_gate\.weight$=q6_k
+^blk\.32\.ffn_up\.weight$=q6_k
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q6_k
+^blk\.33\.attn_q\.weight$=q6_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q6_k
+^blk\.33\.ffn_gate\.weight$=q6_k
+^blk\.33\.ffn_up\.weight$=q6_k
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q6_k
+^blk\.34\.attn_q\.weight$=q6_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q6_k
+^blk\.34\.ffn_gate\.weight$=q6_k
+^blk\.34\.ffn_up\.weight$=q6_k
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q6_k
+^blk\.35\.attn_q\.weight$=q6_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q6_k
+^blk\.35\.ffn_gate\.weight$=q6_k
+^blk\.35\.ffn_up\.weight$=q6_k
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q6_k
+^blk\.36\.attn_q\.weight$=q6_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q6_k
+^blk\.36\.ffn_gate\.weight$=q6_k
+^blk\.36\.ffn_up\.weight$=q6_k
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q6_k
+^blk\.37\.attn_q\.weight$=q6_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q6_k
+^blk\.37\.ffn_gate\.weight$=q6_k
+^blk\.37\.ffn_up\.weight$=q6_k
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q6_k
+^blk\.38\.attn_q\.weight$=q6_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q6_k
+^blk\.38\.ffn_gate\.weight$=q6_k
+^blk\.38\.ffn_up\.weight$=q6_k
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q8_0
+^blk\.39\.attn_q\.weight$=q6_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q8_0
+^blk\.39\.ffn_gate\.weight$=q8_0
+^blk\.39\.ffn_up\.weight$=q8_0

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K_L.layout.json

@@ -0,0 +1,85 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q6_K_L",
+  "base_type": "q6_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.50",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "L",
+  "rung_share": 0.5,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q6_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 12569166000,
+    "map_body+embd": 13502439600,
+    "ratio": 1.0743,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.501,
+  "body_bpw": 7.404,
+  "body_bpw_heuristic": 6.839,
+  "total_params": 14765593600,
+  "file_bytes_pred": 13504139440,
+  "file_bpw_pred": 7.317,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q8_0": 31,
+      "q6_k": 9
+    },
+    "attn_q": {
+      "q6_k": 38,
+      "q8_0": 2
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q8_0": 20,
+      "q6_k": 20
+    },
+    "ffn_gate": {
+      "q6_k": 30,
+      "q8_0": 10
+    },
+    "ffn_up": {
+      "q8_0": 20,
+      "q6_k": 20
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q6_K_L",
+  "quantize_ftype": "Q6_K",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K_L.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q6_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q8_0
+^blk\.0\.attn_q\.weight$=q6_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q8_0
+^blk\.0\.ffn_gate\.weight$=q6_k
+^blk\.0\.ffn_up\.weight$=q8_0
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q6_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q8_0
+^blk\.1\.ffn_gate\.weight$=q8_0
+^blk\.1\.ffn_up\.weight$=q8_0
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q6_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q8_0
+^blk\.2\.ffn_gate\.weight$=q8_0
+^blk\.2\.ffn_up\.weight$=q8_0
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q6_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q8_0
+^blk\.3\.ffn_gate\.weight$=q8_0
+^blk\.3\.ffn_up\.weight$=q8_0
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q6_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q8_0
+^blk\.4\.ffn_gate\.weight$=q8_0
+^blk\.4\.ffn_up\.weight$=q8_0
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q6_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q8_0
+^blk\.5\.ffn_gate\.weight$=q8_0
+^blk\.5\.ffn_up\.weight$=q8_0
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q6_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q8_0
+^blk\.6\.ffn_gate\.weight$=q8_0
+^blk\.6\.ffn_up\.weight$=q8_0
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q6_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q8_0
+^blk\.7\.ffn_gate\.weight$=q8_0
+^blk\.7\.ffn_up\.weight$=q8_0
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q6_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q8_0
+^blk\.8\.ffn_gate\.weight$=q8_0
+^blk\.8\.ffn_up\.weight$=q8_0
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q6_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q8_0
+^blk\.9\.ffn_gate\.weight$=q8_0
+^blk\.9\.ffn_up\.weight$=q8_0
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q8_0
+^blk\.10\.attn_q\.weight$=q8_0
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q6_k
+^blk\.10\.ffn_gate\.weight$=q6_k
+^blk\.10\.ffn_up\.weight$=q6_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q8_0
+^blk\.11\.attn_q\.weight$=q8_0
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q6_k
+^blk\.11\.ffn_gate\.weight$=q6_k
+^blk\.11\.ffn_up\.weight$=q6_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q8_0
+^blk\.12\.attn_q\.weight$=q6_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q6_k
+^blk\.12\.ffn_gate\.weight$=q6_k
+^blk\.12\.ffn_up\.weight$=q6_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q8_0
+^blk\.13\.attn_q\.weight$=q6_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q6_k
+^blk\.13\.ffn_gate\.weight$=q6_k
+^blk\.13\.ffn_up\.weight$=q6_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q8_0
+^blk\.14\.attn_q\.weight$=q6_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q6_k
+^blk\.14\.ffn_gate\.weight$=q6_k
+^blk\.14\.ffn_up\.weight$=q6_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q8_0
+^blk\.15\.attn_q\.weight$=q6_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q6_k
+^blk\.15\.ffn_gate\.weight$=q6_k
+^blk\.15\.ffn_up\.weight$=q6_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q8_0
+^blk\.16\.attn_q\.weight$=q6_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q6_k
+^blk\.16\.ffn_gate\.weight$=q6_k
+^blk\.16\.ffn_up\.weight$=q6_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q8_0
+^blk\.17\.attn_q\.weight$=q6_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q6_k
+^blk\.17\.ffn_gate\.weight$=q6_k
+^blk\.17\.ffn_up\.weight$=q6_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q8_0
+^blk\.18\.attn_q\.weight$=q6_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q6_k
+^blk\.18\.ffn_gate\.weight$=q6_k
+^blk\.18\.ffn_up\.weight$=q6_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q8_0
+^blk\.19\.attn_q\.weight$=q6_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q6_k
+^blk\.19\.ffn_gate\.weight$=q6_k
+^blk\.19\.ffn_up\.weight$=q6_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q8_0
+^blk\.20\.attn_q\.weight$=q6_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q6_k
+^blk\.20\.ffn_gate\.weight$=q6_k
+^blk\.20\.ffn_up\.weight$=q6_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q8_0
+^blk\.21\.attn_q\.weight$=q6_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q6_k
+^blk\.21\.ffn_gate\.weight$=q6_k
+^blk\.21\.ffn_up\.weight$=q6_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q8_0
+^blk\.22\.attn_q\.weight$=q6_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q6_k
+^blk\.22\.ffn_gate\.weight$=q6_k
+^blk\.22\.ffn_up\.weight$=q6_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q8_0
+^blk\.23\.attn_q\.weight$=q6_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q6_k
+^blk\.23\.ffn_gate\.weight$=q6_k
+^blk\.23\.ffn_up\.weight$=q6_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q8_0
+^blk\.24\.attn_q\.weight$=q6_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q6_k
+^blk\.24\.ffn_gate\.weight$=q6_k
+^blk\.24\.ffn_up\.weight$=q6_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q8_0
+^blk\.25\.attn_q\.weight$=q6_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q6_k
+^blk\.25\.ffn_gate\.weight$=q6_k
+^blk\.25\.ffn_up\.weight$=q6_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q8_0
+^blk\.26\.attn_q\.weight$=q6_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q6_k
+^blk\.26\.ffn_gate\.weight$=q6_k
+^blk\.26\.ffn_up\.weight$=q6_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q8_0
+^blk\.27\.attn_q\.weight$=q6_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q6_k
+^blk\.27\.ffn_gate\.weight$=q6_k
+^blk\.27\.ffn_up\.weight$=q6_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q8_0
+^blk\.28\.attn_q\.weight$=q6_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q6_k
+^blk\.28\.ffn_gate\.weight$=q6_k
+^blk\.28\.ffn_up\.weight$=q6_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q8_0
+^blk\.29\.attn_q\.weight$=q6_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q6_k
+^blk\.29\.ffn_gate\.weight$=q6_k
+^blk\.29\.ffn_up\.weight$=q6_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q6_k
+^blk\.30\.attn_q\.weight$=q6_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q8_0
+^blk\.30\.ffn_gate\.weight$=q6_k
+^blk\.30\.ffn_up\.weight$=q8_0
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q6_k
+^blk\.31\.attn_q\.weight$=q6_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q8_0
+^blk\.31\.ffn_gate\.weight$=q6_k
+^blk\.31\.ffn_up\.weight$=q8_0
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q6_k
+^blk\.32\.attn_q\.weight$=q6_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q8_0
+^blk\.32\.ffn_gate\.weight$=q6_k
+^blk\.32\.ffn_up\.weight$=q8_0
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q6_k
+^blk\.33\.attn_q\.weight$=q6_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q8_0
+^blk\.33\.ffn_gate\.weight$=q6_k
+^blk\.33\.ffn_up\.weight$=q8_0
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q6_k
+^blk\.34\.attn_q\.weight$=q6_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q8_0
+^blk\.34\.ffn_gate\.weight$=q6_k
+^blk\.34\.ffn_up\.weight$=q8_0
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q6_k
+^blk\.35\.attn_q\.weight$=q6_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q8_0
+^blk\.35\.ffn_gate\.weight$=q6_k
+^blk\.35\.ffn_up\.weight$=q8_0
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q6_k
+^blk\.36\.attn_q\.weight$=q6_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q8_0
+^blk\.36\.ffn_gate\.weight$=q6_k
+^blk\.36\.ffn_up\.weight$=q8_0
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q6_k
+^blk\.37\.attn_q\.weight$=q6_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q8_0
+^blk\.37\.ffn_gate\.weight$=q6_k
+^blk\.37\.ffn_up\.weight$=q8_0
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q6_k
+^blk\.38\.attn_q\.weight$=q6_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q8_0
+^blk\.38\.ffn_gate\.weight$=q6_k
+^blk\.38\.ffn_up\.weight$=q8_0
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q8_0
+^blk\.39\.attn_q\.weight$=q6_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q8_0
+^blk\.39\.ffn_gate\.weight$=q8_0
+^blk\.39\.ffn_up\.weight$=q8_0

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K_S.layout.json

@@ -0,0 +1,82 @@
+{
+  "model": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-bf16.gguf",
+  "imatrix": "/media/m2teamgroup/models/HybridAgent-14B-Agentic-GGUF/togethercomputer_HybridAgent-14B-Agentic-imatrix.gguf",
+  "ftype": "Q6_K_S",
+  "base_type": "q6_k",
+  "family": "k",
+  "recipe": "ident",
+  "variant": "shr0.90",
+  "auto_crush": false,
+  "rung_mode": "share",
+  "tier": "S",
+  "rung_share": 0.9,
+  "crush_pass_bytes": null,
+  "embd_shares": {
+    "output.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    },
+    "token_embd.weight": {
+      "share": 0.0507,
+      "type": "q6_k"
+    }
+  },
+  "arch": "dense",
+  "tied": false,
+  "n_layer": 40,
+  "body_rule": true,
+  "pins": {},
+  "shape_pins": {},
+  "fixed_pins": {},
+  "embeddings": {
+    "output.weight": "q6_k",
+    "token_embd.weight": "q6_k"
+  },
+  "bytes": {
+    "heuristic_body+embd": 12569166000,
+    "map_body+embd": 12362195120,
+    "ratio": 0.9835,
+    "note": "quantizable, unpinned tensors only; pinned/unquantized tensors are identical in both"
+  },
+  "body_base_share": 0.901,
+  "body_bpw": 6.714,
+  "body_bpw_heuristic": 6.839,
+  "total_params": 14765593600,
+  "file_bytes_pred": 12363894960,
+  "file_bpw_pred": 6.699,
+  "histogram": {
+    "attn_k": {
+      "q8_0": 40
+    },
+    "attn_output": {
+      "q6_k": 20,
+      "q8_0": 20
+    },
+    "attn_q": {
+      "q6_k": 40
+    },
+    "attn_v": {
+      "q8_0": 40
+    },
+    "ffn_down": {
+      "q6_k": 40
+    },
+    "ffn_gate": {
+      "q6_k": 40
+    },
+    "ffn_up": {
+      "q6_k": 39,
+      "q8_0": 1
+    }
+  },
+  "prior": "layout/prior.json",
+  "pin_rule": "floor",
+  "embd_cap": "heur",
+  "generator": "auto_quant_v2.layout",
+  "generator_key": "3bf8b43e20a68f0b",
+  "quant_type": "Q6_K_S",
+  "quantize_ftype": "Q6_K",
+  "llama_cpp_version": "b11716",
+  "unknown_kinds": {},
+  "commit": "2e3efd54e5964c13d80f52925ef8ce2003b86465"
+}

layouts/togethercomputer_HybridAgent-14B-Agentic-Q6_K_S.tensor-types.txt

@@ -0,0 +1,282 @@
+^output\.weight$=q6_k
+^token_embd\.weight$=q6_k
+^blk\.0\.attn_k\.weight$=q8_0
+^blk\.0\.attn_output\.weight$=q6_k
+^blk\.0\.attn_q\.weight$=q6_k
+^blk\.0\.attn_v\.weight$=q8_0
+^blk\.0\.ffn_down\.weight$=q6_k
+^blk\.0\.ffn_gate\.weight$=q6_k
+^blk\.0\.ffn_up\.weight$=q6_k
+^blk\.1\.attn_k\.weight$=q8_0
+^blk\.1\.attn_output\.weight$=q8_0
+^blk\.1\.attn_q\.weight$=q6_k
+^blk\.1\.attn_v\.weight$=q8_0
+^blk\.1\.ffn_down\.weight$=q6_k
+^blk\.1\.ffn_gate\.weight$=q6_k
+^blk\.1\.ffn_up\.weight$=q6_k
+^blk\.2\.attn_k\.weight$=q8_0
+^blk\.2\.attn_output\.weight$=q8_0
+^blk\.2\.attn_q\.weight$=q6_k
+^blk\.2\.attn_v\.weight$=q8_0
+^blk\.2\.ffn_down\.weight$=q6_k
+^blk\.2\.ffn_gate\.weight$=q6_k
+^blk\.2\.ffn_up\.weight$=q6_k
+^blk\.3\.attn_k\.weight$=q8_0
+^blk\.3\.attn_output\.weight$=q8_0
+^blk\.3\.attn_q\.weight$=q6_k
+^blk\.3\.attn_v\.weight$=q8_0
+^blk\.3\.ffn_down\.weight$=q6_k
+^blk\.3\.ffn_gate\.weight$=q6_k
+^blk\.3\.ffn_up\.weight$=q6_k
+^blk\.4\.attn_k\.weight$=q8_0
+^blk\.4\.attn_output\.weight$=q8_0
+^blk\.4\.attn_q\.weight$=q6_k
+^blk\.4\.attn_v\.weight$=q8_0
+^blk\.4\.ffn_down\.weight$=q6_k
+^blk\.4\.ffn_gate\.weight$=q6_k
+^blk\.4\.ffn_up\.weight$=q6_k
+^blk\.5\.attn_k\.weight$=q8_0
+^blk\.5\.attn_output\.weight$=q8_0
+^blk\.5\.attn_q\.weight$=q6_k
+^blk\.5\.attn_v\.weight$=q8_0
+^blk\.5\.ffn_down\.weight$=q6_k
+^blk\.5\.ffn_gate\.weight$=q6_k
+^blk\.5\.ffn_up\.weight$=q6_k
+^blk\.6\.attn_k\.weight$=q8_0
+^blk\.6\.attn_output\.weight$=q8_0
+^blk\.6\.attn_q\.weight$=q6_k
+^blk\.6\.attn_v\.weight$=q8_0
+^blk\.6\.ffn_down\.weight$=q6_k
+^blk\.6\.ffn_gate\.weight$=q6_k
+^blk\.6\.ffn_up\.weight$=q6_k
+^blk\.7\.attn_k\.weight$=q8_0
+^blk\.7\.attn_output\.weight$=q8_0
+^blk\.7\.attn_q\.weight$=q6_k
+^blk\.7\.attn_v\.weight$=q8_0
+^blk\.7\.ffn_down\.weight$=q6_k
+^blk\.7\.ffn_gate\.weight$=q6_k
+^blk\.7\.ffn_up\.weight$=q6_k
+^blk\.8\.attn_k\.weight$=q8_0
+^blk\.8\.attn_output\.weight$=q8_0
+^blk\.8\.attn_q\.weight$=q6_k
+^blk\.8\.attn_v\.weight$=q8_0
+^blk\.8\.ffn_down\.weight$=q6_k
+^blk\.8\.ffn_gate\.weight$=q6_k
+^blk\.8\.ffn_up\.weight$=q6_k
+^blk\.9\.attn_k\.weight$=q8_0
+^blk\.9\.attn_output\.weight$=q8_0
+^blk\.9\.attn_q\.weight$=q6_k
+^blk\.9\.attn_v\.weight$=q8_0
+^blk\.9\.ffn_down\.weight$=q6_k
+^blk\.9\.ffn_gate\.weight$=q6_k
+^blk\.9\.ffn_up\.weight$=q6_k
+^blk\.10\.attn_k\.weight$=q8_0
+^blk\.10\.attn_output\.weight$=q8_0
+^blk\.10\.attn_q\.weight$=q6_k
+^blk\.10\.attn_v\.weight$=q8_0
+^blk\.10\.ffn_down\.weight$=q6_k
+^blk\.10\.ffn_gate\.weight$=q6_k
+^blk\.10\.ffn_up\.weight$=q6_k
+^blk\.11\.attn_k\.weight$=q8_0
+^blk\.11\.attn_output\.weight$=q8_0
+^blk\.11\.attn_q\.weight$=q6_k
+^blk\.11\.attn_v\.weight$=q8_0
+^blk\.11\.ffn_down\.weight$=q6_k
+^blk\.11\.ffn_gate\.weight$=q6_k
+^blk\.11\.ffn_up\.weight$=q6_k
+^blk\.12\.attn_k\.weight$=q8_0
+^blk\.12\.attn_output\.weight$=q8_0
+^blk\.12\.attn_q\.weight$=q6_k
+^blk\.12\.attn_v\.weight$=q8_0
+^blk\.12\.ffn_down\.weight$=q6_k
+^blk\.12\.ffn_gate\.weight$=q6_k
+^blk\.12\.ffn_up\.weight$=q6_k
+^blk\.13\.attn_k\.weight$=q8_0
+^blk\.13\.attn_output\.weight$=q8_0
+^blk\.13\.attn_q\.weight$=q6_k
+^blk\.13\.attn_v\.weight$=q8_0
+^blk\.13\.ffn_down\.weight$=q6_k
+^blk\.13\.ffn_gate\.weight$=q6_k
+^blk\.13\.ffn_up\.weight$=q6_k
+^blk\.14\.attn_k\.weight$=q8_0
+^blk\.14\.attn_output\.weight$=q8_0
+^blk\.14\.attn_q\.weight$=q6_k
+^blk\.14\.attn_v\.weight$=q8_0
+^blk\.14\.ffn_down\.weight$=q6_k
+^blk\.14\.ffn_gate\.weight$=q6_k
+^blk\.14\.ffn_up\.weight$=q6_k
+^blk\.15\.attn_k\.weight$=q8_0
+^blk\.15\.attn_output\.weight$=q8_0
+^blk\.15\.attn_q\.weight$=q6_k
+^blk\.15\.attn_v\.weight$=q8_0
+^blk\.15\.ffn_down\.weight$=q6_k
+^blk\.15\.ffn_gate\.weight$=q6_k
+^blk\.15\.ffn_up\.weight$=q6_k
+^blk\.16\.attn_k\.weight$=q8_0
+^blk\.16\.attn_output\.weight$=q8_0
+^blk\.16\.attn_q\.weight$=q6_k
+^blk\.16\.attn_v\.weight$=q8_0
+^blk\.16\.ffn_down\.weight$=q6_k
+^blk\.16\.ffn_gate\.weight$=q6_k
+^blk\.16\.ffn_up\.weight$=q6_k
+^blk\.17\.attn_k\.weight$=q8_0
+^blk\.17\.attn_output\.weight$=q8_0
+^blk\.17\.attn_q\.weight$=q6_k
+^blk\.17\.attn_v\.weight$=q8_0
+^blk\.17\.ffn_down\.weight$=q6_k
+^blk\.17\.ffn_gate\.weight$=q6_k
+^blk\.17\.ffn_up\.weight$=q6_k
+^blk\.18\.attn_k\.weight$=q8_0
+^blk\.18\.attn_output\.weight$=q8_0
+^blk\.18\.attn_q\.weight$=q6_k
+^blk\.18\.attn_v\.weight$=q8_0
+^blk\.18\.ffn_down\.weight$=q6_k
+^blk\.18\.ffn_gate\.weight$=q6_k
+^blk\.18\.ffn_up\.weight$=q6_k
+^blk\.19\.attn_k\.weight$=q8_0
+^blk\.19\.attn_output\.weight$=q8_0
+^blk\.19\.attn_q\.weight$=q6_k
+^blk\.19\.attn_v\.weight$=q8_0
+^blk\.19\.ffn_down\.weight$=q6_k
+^blk\.19\.ffn_gate\.weight$=q6_k
+^blk\.19\.ffn_up\.weight$=q6_k
+^blk\.20\.attn_k\.weight$=q8_0
+^blk\.20\.attn_output\.weight$=q8_0
+^blk\.20\.attn_q\.weight$=q6_k
+^blk\.20\.attn_v\.weight$=q8_0
+^blk\.20\.ffn_down\.weight$=q6_k
+^blk\.20\.ffn_gate\.weight$=q6_k
+^blk\.20\.ffn_up\.weight$=q6_k
+^blk\.21\.attn_k\.weight$=q8_0
+^blk\.21\.attn_output\.weight$=q6_k
+^blk\.21\.attn_q\.weight$=q6_k
+^blk\.21\.attn_v\.weight$=q8_0
+^blk\.21\.ffn_down\.weight$=q6_k
+^blk\.21\.ffn_gate\.weight$=q6_k
+^blk\.21\.ffn_up\.weight$=q6_k
+^blk\.22\.attn_k\.weight$=q8_0
+^blk\.22\.attn_output\.weight$=q6_k
+^blk\.22\.attn_q\.weight$=q6_k
+^blk\.22\.attn_v\.weight$=q8_0
+^blk\.22\.ffn_down\.weight$=q6_k
+^blk\.22\.ffn_gate\.weight$=q6_k
+^blk\.22\.ffn_up\.weight$=q6_k
+^blk\.23\.attn_k\.weight$=q8_0
+^blk\.23\.attn_output\.weight$=q6_k
+^blk\.23\.attn_q\.weight$=q6_k
+^blk\.23\.attn_v\.weight$=q8_0
+^blk\.23\.ffn_down\.weight$=q6_k
+^blk\.23\.ffn_gate\.weight$=q6_k
+^blk\.23\.ffn_up\.weight$=q6_k
+^blk\.24\.attn_k\.weight$=q8_0
+^blk\.24\.attn_output\.weight$=q6_k
+^blk\.24\.attn_q\.weight$=q6_k
+^blk\.24\.attn_v\.weight$=q8_0
+^blk\.24\.ffn_down\.weight$=q6_k
+^blk\.24\.ffn_gate\.weight$=q6_k
+^blk\.24\.ffn_up\.weight$=q6_k
+^blk\.25\.attn_k\.weight$=q8_0
+^blk\.25\.attn_output\.weight$=q6_k
+^blk\.25\.attn_q\.weight$=q6_k
+^blk\.25\.attn_v\.weight$=q8_0
+^blk\.25\.ffn_down\.weight$=q6_k
+^blk\.25\.ffn_gate\.weight$=q6_k
+^blk\.25\.ffn_up\.weight$=q6_k
+^blk\.26\.attn_k\.weight$=q8_0
+^blk\.26\.attn_output\.weight$=q6_k
+^blk\.26\.attn_q\.weight$=q6_k
+^blk\.26\.attn_v\.weight$=q8_0
+^blk\.26\.ffn_down\.weight$=q6_k
+^blk\.26\.ffn_gate\.weight$=q6_k
+^blk\.26\.ffn_up\.weight$=q6_k
+^blk\.27\.attn_k\.weight$=q8_0
+^blk\.27\.attn_output\.weight$=q6_k
+^blk\.27\.attn_q\.weight$=q6_k
+^blk\.27\.attn_v\.weight$=q8_0
+^blk\.27\.ffn_down\.weight$=q6_k
+^blk\.27\.ffn_gate\.weight$=q6_k
+^blk\.27\.ffn_up\.weight$=q6_k
+^blk\.28\.attn_k\.weight$=q8_0
+^blk\.28\.attn_output\.weight$=q6_k
+^blk\.28\.attn_q\.weight$=q6_k
+^blk\.28\.attn_v\.weight$=q8_0
+^blk\.28\.ffn_down\.weight$=q6_k
+^blk\.28\.ffn_gate\.weight$=q6_k
+^blk\.28\.ffn_up\.weight$=q6_k
+^blk\.29\.attn_k\.weight$=q8_0
+^blk\.29\.attn_output\.weight$=q6_k
+^blk\.29\.attn_q\.weight$=q6_k
+^blk\.29\.attn_v\.weight$=q8_0
+^blk\.29\.ffn_down\.weight$=q6_k
+^blk\.29\.ffn_gate\.weight$=q6_k
+^blk\.29\.ffn_up\.weight$=q6_k
+^blk\.30\.attn_k\.weight$=q8_0
+^blk\.30\.attn_output\.weight$=q6_k
+^blk\.30\.attn_q\.weight$=q6_k
+^blk\.30\.attn_v\.weight$=q8_0
+^blk\.30\.ffn_down\.weight$=q6_k
+^blk\.30\.ffn_gate\.weight$=q6_k
+^blk\.30\.ffn_up\.weight$=q6_k
+^blk\.31\.attn_k\.weight$=q8_0
+^blk\.31\.attn_output\.weight$=q6_k
+^blk\.31\.attn_q\.weight$=q6_k
+^blk\.31\.attn_v\.weight$=q8_0
+^blk\.31\.ffn_down\.weight$=q6_k
+^blk\.31\.ffn_gate\.weight$=q6_k
+^blk\.31\.ffn_up\.weight$=q6_k
+^blk\.32\.attn_k\.weight$=q8_0
+^blk\.32\.attn_output\.weight$=q6_k
+^blk\.32\.attn_q\.weight$=q6_k
+^blk\.32\.attn_v\.weight$=q8_0
+^blk\.32\.ffn_down\.weight$=q6_k
+^blk\.32\.ffn_gate\.weight$=q6_k
+^blk\.32\.ffn_up\.weight$=q6_k
+^blk\.33\.attn_k\.weight$=q8_0
+^blk\.33\.attn_output\.weight$=q6_k
+^blk\.33\.attn_q\.weight$=q6_k
+^blk\.33\.attn_v\.weight$=q8_0
+^blk\.33\.ffn_down\.weight$=q6_k
+^blk\.33\.ffn_gate\.weight$=q6_k
+^blk\.33\.ffn_up\.weight$=q6_k
+^blk\.34\.attn_k\.weight$=q8_0
+^blk\.34\.attn_output\.weight$=q6_k
+^blk\.34\.attn_q\.weight$=q6_k
+^blk\.34\.attn_v\.weight$=q8_0
+^blk\.34\.ffn_down\.weight$=q6_k
+^blk\.34\.ffn_gate\.weight$=q6_k
+^blk\.34\.ffn_up\.weight$=q6_k
+^blk\.35\.attn_k\.weight$=q8_0
+^blk\.35\.attn_output\.weight$=q6_k
+^blk\.35\.attn_q\.weight$=q6_k
+^blk\.35\.attn_v\.weight$=q8_0
+^blk\.35\.ffn_down\.weight$=q6_k
+^blk\.35\.ffn_gate\.weight$=q6_k
+^blk\.35\.ffn_up\.weight$=q6_k
+^blk\.36\.attn_k\.weight$=q8_0
+^blk\.36\.attn_output\.weight$=q6_k
+^blk\.36\.attn_q\.weight$=q6_k
+^blk\.36\.attn_v\.weight$=q8_0
+^blk\.36\.ffn_down\.weight$=q6_k
+^blk\.36\.ffn_gate\.weight$=q6_k
+^blk\.36\.ffn_up\.weight$=q6_k
+^blk\.37\.attn_k\.weight$=q8_0
+^blk\.37\.attn_output\.weight$=q6_k
+^blk\.37\.attn_q\.weight$=q6_k
+^blk\.37\.attn_v\.weight$=q8_0
+^blk\.37\.ffn_down\.weight$=q6_k
+^blk\.37\.ffn_gate\.weight$=q6_k
+^blk\.37\.ffn_up\.weight$=q6_k
+^blk\.38\.attn_k\.weight$=q8_0
+^blk\.38\.attn_output\.weight$=q6_k
+^blk\.38\.attn_q\.weight$=q6_k
+^blk\.38\.attn_v\.weight$=q8_0
+^blk\.38\.ffn_down\.weight$=q6_k
+^blk\.38\.ffn_gate\.weight$=q6_k
+^blk\.38\.ffn_up\.weight$=q6_k
+^blk\.39\.attn_k\.weight$=q8_0
+^blk\.39\.attn_output\.weight$=q6_k
+^blk\.39\.attn_q\.weight$=q6_k
+^blk\.39\.attn_v\.weight$=q8_0
+^blk\.39\.ffn_down\.weight$=q6_k
+^blk\.39\.ffn_gate\.weight$=q6_k
+^blk\.39\.ffn_up\.weight$=q8_0