INTERPRETIC commited on
Commit
3c348d7
·
verified ·
1 Parent(s): dfe0112

Upload quantized weights and model card

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ license_link: https://www.apache.org/licenses/LICENSE-2.0
4
+ library_name: mlx
5
+ pipeline_tag: image-text-to-text
6
+ base_model:
7
+ - thinkingmachines/Inkling-2-Mini
8
+ tags:
9
+ - inkling
10
+ - moe
11
+ - mlx
12
+ ---
13
+
14
+ # Inkling-2-Mini-MLX-4bit
15
+
16
+ [Inkling 2 Mini](https://huggingface.co/thinkingmachines/Inkling-2-Mini) is a multimodal looped mixture-of-experts model from Thinking Machines Lab: 26.4B parameters, 3.4B active (5.4B effective with three passes through the looped middle layers), text and image input, up to 1M tokens of context. This repository is an unmodified quantization of the official checkpoint (revision `1fb4134`).
17
+
18
+ MLX build for Apple Silicon (MLX, oMLX, LM Studio). Runs on a Mac with 32 GB of unified memory or more.
19
+
20
+ Other formats from INTERPRETIC AI: [GGUF Q4_K_M](https://huggingface.co/InterpreticAI/Inkling-2-Mini-GGUF) · [GGUF IQ3_XXS](https://huggingface.co/InterpreticAI/Inkling-2-Mini-IQ3_XXS-GGUF) · [MLX 4-bit](https://huggingface.co/InterpreticAI/Inkling-2-Mini-MLX-4bit)
21
+
22
+ ## Files
23
+
24
+ Four `model-0000N-of-00004.safetensors` shards (17 GB in total), the configuration and the tokenizer. The vision tower is included in bfloat16. The routed experts keep the original NVFP4 quantization. Attention, embedding and output weights are 4-bit (group size 64). SHA-256 checksums are listed in `SHA256SUMS`.
25
+
26
+ ## Usage
27
+
28
+ Inkling 2 Mini support in mlx-lm is in [ml-explore/mlx-lm#2086](https://github.com/ml-explore/mlx-lm/pull/2086) and not yet in a release.
29
+
30
+ ```bash
31
+ pip install -U "git+https://github.com/ml-explore/mlx-lm.git@refs/pull/2086/head"
32
+ mlx_lm.generate --model InterpreticAI/Inkling-2-Mini-MLX-4bit --prompt "Hello" --max-tokens 512
33
+ ```
34
+
35
+ Tested with text input. Image input through mlx-vlm has not been tested.
36
+
37
+ ## Notes
38
+
39
+ - Inkling 2 Mini is a thinking model: it writes its reasoning before the answer. Use a chat client or server that applies the model's chat template.
40
+ - Quality checks here are spot checks (English, Chinese, code and one image), not a full benchmark.
41
+ - Support for this architecture is new in all runtimes. Use a recent build.
42
+
43
+ ## License
44
+
45
+ Apache 2.0, as the original model. See the [model card of Thinking Machines Lab](https://huggingface.co/thinkingmachines/Inkling-2-Mini) and its [acceptable use policy](https://thinkingmachines.ai/model-acceptable-use-policy/).
46
+
47
+ INTERPRETIC AI · [interpretic.ai](https://interpretic.ai)
SHA256SUMS ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 4d8b6bb2ec5c459168a90e9b45f36866d66888218818fe88f21bc3f812fc24fc README.md
2
+ 53a9035f3255945704b5857657cfb00c1165090c0641c4f782a0cbf268cc16dc chat_template.jinja
3
+ 4fd8ce2475949692a3563ab81278b9294ae4c3eae7d9b3c168e37056b68263d9 config.json
4
+ a08737c07a65a8fdf69f7cf086576210e5aad4b241437090b26d7b0c4f7905f7 generation_config.json
5
+ acc233d29fdf2762fdc0edfb46e229be4a935611847e1105435f13a9ae7c0894 model-00001-of-00004.safetensors
6
+ 717b9ae80a043710944e3773a2ff76655f3b788d50ac88c1934bea1b05ee8c83 model-00002-of-00004.safetensors
7
+ 1f42cada3567f4613a971244f3bec5f1083d3ae3aca933512ae49643a6748139 model-00003-of-00004.safetensors
8
+ 0aeeff012da96da5d5e6bf6f0b73a54de4c94bbc1c6321586cb4565469a02684 model-00004-of-00004.safetensors
9
+ 6e2d0fbb83f4ec2488f22f521e57855375ea93b473a704da10871398f952d9f5 model.safetensors.index.json
10
+ 7f332727f9a3f093dd452a39cc978e6757627e5dad1b0673aa4e75174f236dd6 tokenizer.json
11
+ 2090107583ef1cd15a08cb644e3218dafe62f29dbadf5c8026839d13b8b24dde tokenizer_config.json
chat_template.jinja ADDED
@@ -0,0 +1,137 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set effort_map = {"none": 0.0, "minimal": 0.1, "low": 0.2, "medium": 0.7, "high": 0.9, "xhigh": 0.99, "max": 0.99} -%}
2
+ {%- set role_token = {"user": "<|message_user|>", "assistant": "<|message_model|>", "system": "<|message_system|>", "tool": "<|message_tool|>"} -%}
3
+
4
+ {%- macro emit_thinking_effort() -%}
5
+ {%- set eff = reasoning_effort if reasoning_effort is defined and reasoning_effort is not none else 0.9 -%}
6
+ {%- if eff is string -%}
7
+ {%- set key = eff | trim -%}
8
+ {%- if key not in effort_map -%}
9
+ {{- raise_exception("Unknown reasoning_effort: " ~ eff) -}}
10
+ {%- endif -%}
11
+ {%- set num = effort_map[key] -%}
12
+ {%- else -%}
13
+ {%- set num = eff | float -%}
14
+ {%- endif -%}
15
+ {%- if num < 0.0 or num > 0.99 -%}
16
+ {{- raise_exception("reasoning_effort must be in [0.0, 0.99]") -}}
17
+ {%- endif -%}
18
+ {{- "<|message_system|><|content_text|>Thinking effort level: " -}}
19
+ {%- if num == 0.0 -%}0{%- else -%}{{ num }}{%- endif -%}
20
+ {{- "<|end_message|>" -}}
21
+ {%- endmacro -%}
22
+
23
+ {%- if tools -%}
24
+ {%- set tool_state = namespace(specs=[]) -%}
25
+ {%- for tool in tools -%}
26
+ {%- set fn = tool.function if tool.function is defined else tool -%}
27
+ {%- set spec = {
28
+ "description": (fn.description if fn.description is defined and fn.description else ""),
29
+ "name": fn.name,
30
+ "parameters": (fn.parameters if fn.parameters is defined and fn.parameters else {}),
31
+ "type": (tool.type if tool.type is defined and tool.type else "function"),
32
+ } -%}
33
+ {%- set tool_state.specs = tool_state.specs + [spec] -%}
34
+ {%- endfor -%}
35
+ {{- "<|message_system|>tool_declare<|content_xml|>" -}}
36
+ {{- tool_state.specs | tojson(sort_keys=true, separators=(",", ":")) -}}
37
+ {{- "<|end_message|>" -}}
38
+ {%- endif -%}
39
+
40
+ {%- set state = namespace(effort_emitted=false) -%}
41
+ {%- for message in messages -%}
42
+ {%- if message.role not in role_token -%}
43
+ {{- raise_exception("Unknown message role: " ~ message.role) -}}
44
+ {%- endif -%}
45
+ {%- if not state.effort_emitted and message.role != "system" -%}
46
+ {{- emit_thinking_effort() -}}
47
+ {%- set state.effort_emitted = true -%}
48
+ {%- endif -%}
49
+
50
+ {%- set header = namespace(text=role_token[message.role]) -%}
51
+ {%- if message.role == "tool" -%}
52
+ {%- if message.name is defined and message.name -%}
53
+ {%- set header.text = header.text ~ message.name -%}
54
+ {%- elif message.tool_call_id is defined and message.tool_call_id -%}
55
+ {%- for prev in messages -%}
56
+ {%- if prev.role == "assistant" and prev.tool_calls -%}
57
+ {%- for tc in prev.tool_calls -%}
58
+ {%- if tc.id is defined and tc.id == message.tool_call_id and tc.function.name is defined -%}
59
+ {%- set header.text = header.text ~ tc.function.name -%}
60
+ {%- endif -%}
61
+ {%- endfor -%}
62
+ {%- endif -%}
63
+ {%- endfor -%}
64
+ {%- endif -%}
65
+ {%- elif message.role == "assistant" and message.name is defined and message.name -%}
66
+ {%- set header.text = header.text ~ message.name -%}
67
+ {%- endif -%}
68
+ {%- set rtok = header.text -%}
69
+ {%- set rendered -%}
70
+
71
+ {%- if message.role == "assistant" and message.reasoning_content is defined and message.reasoning_content -%}
72
+ {{- rtok ~ "<|content_thinking|>" ~ message.reasoning_content ~ "<|end_message|>" -}}
73
+ {%- endif -%}
74
+
75
+ {%- if message.content is string and (message.content or not (message.role == "assistant" and (message.tool_calls or message.reasoning_content))) -%}
76
+ {{- rtok ~ "<|content_text|>" ~ message.content ~ "<|end_message|>" -}}
77
+ {%- elif message.content -%}
78
+ {%- for part in message.content -%}
79
+ {%- if part is string -%}
80
+ {{- rtok ~ "<|content_text|>" ~ part ~ "<|end_message|>" -}}
81
+ {%- elif part.type is not defined or part.type in ("text", "input_text") -%}
82
+ {%- set text_part = (part.text if part.text is defined and part.text is string else "") -%}
83
+ {{- rtok ~ "<|content_text|>" ~ text_part ~ "<|end_message|>" -}}
84
+ {%- elif part.type in ("image", "input_image", "image_url") -%}
85
+ {%- if message.role not in ("user", "tool") -%}
86
+ {{- raise_exception("Image content is only supported for user and tool messages") -}}
87
+ {%- endif -%}
88
+ {{- rtok ~ "<|content_image|><|unused_200054|><|end_message|>" -}}
89
+ {%- elif part.type in ("audio", "input_audio", "audio_url") -%}
90
+ {{- raise_exception("Inkling 2 does not accept audio input") -}}
91
+ {%- else -%}
92
+ {{- raise_exception("Unsupported content part type: " ~ part.type) -}}
93
+ {%- endif -%}
94
+ {%- endfor -%}
95
+ {%- endif -%}
96
+
97
+ {%- if message.role == "assistant" and message.tool_calls -%}
98
+ {%- for tc in message.tool_calls -%}
99
+ {%- set fn = tc.function -%}
100
+ {%- if fn.name is not defined or fn.name is not string -%}
101
+ {{- raise_exception("tool call function name must be a string") -}}
102
+ {%- endif -%}
103
+ {%- if message.name is defined and message.name and message.name != fn.name -%}
104
+ {{- raise_exception("assistant name " ~ message.name ~ " does not match its tool call " ~ fn.name) -}}
105
+ {%- endif -%}
106
+ {%- set args = fn.arguments if fn.arguments is defined and fn.arguments else {} -%}
107
+ {%- if args is string -%}
108
+ {{- raise_exception("tool call arguments must be a parsed object, not a JSON string; canonicalize upstream") -}}
109
+ {%- endif -%}
110
+ {%- if args is not mapping -%}
111
+ {{- raise_exception("tool call arguments must be an object") -}}
112
+ {%- endif -%}
113
+ {{- "<|message_model|>" ~ fn.name ~ "<|content_invoke_tool_json|>" -}}
114
+ {{- '{"name":' ~ (fn.name | tojson(sort_keys=true, separators=(",", ":"))) ~ ',"args":' -}}
115
+ {{- "{" -}}
116
+ {%- for key, value in args | dictsort(case_sensitive=true) -%}
117
+ {{- (key | tojson) ~ ":" ~ (value | tojson(separators=(",", ":"))) -}}
118
+ {%- if not loop.last -%},{%- endif -%}
119
+ {%- endfor -%}
120
+ {{- "}}<|end_message|>" -}}
121
+ {%- endfor -%}
122
+ {%- endif -%}
123
+
124
+ {%- if message.role == "assistant" -%}
125
+ {{- "<|content_model_end_sampling|>" -}}
126
+ {%- endif -%}
127
+ {%- endset -%}
128
+ {%- if message.role == "assistant" -%}
129
+ {%- generation -%}{{- rendered -}}{%- endgeneration -%}
130
+ {%- else -%}
131
+ {{- rendered -}}
132
+ {%- endif -%}
133
+ {%- endfor -%}
134
+
135
+ {%- if not state.effort_emitted -%}
136
+ {{- emit_thinking_effort() -}}
137
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,1907 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Inkling2LoopForConditionalGeneration"
4
+ ],
5
+ "eos_token_id": 200006,
6
+ "model_type": "inkling2_loop",
7
+ "mtp_config": {
8
+ "hidden_size": 2304,
9
+ "num_hidden_layers": 3,
10
+ "num_attention_heads": 24,
11
+ "num_key_value_heads": 8,
12
+ "head_dim": 128,
13
+ "intermediate_size": 9216,
14
+ "max_position_embeddings": 1048576,
15
+ "torch_dtype": "bfloat16",
16
+ "q_bias": false,
17
+ "o_bias": false,
18
+ "rms_norm_eps": 1e-06,
19
+ "rope_theta": 1000000.0,
20
+ "local_layer_ids": [
21
+ 0,
22
+ 2
23
+ ],
24
+ "sliding_window_size": 1024,
25
+ "block_size": 8,
26
+ "markov_rank": 256,
27
+ "markov_head_type": "rnn_hidden",
28
+ "target_layer_ids": [
29
+ 5,
30
+ 23,
31
+ 41,
32
+ 48
33
+ ]
34
+ },
35
+ "quantization": {
36
+ "group_size": 16,
37
+ "bits": 4,
38
+ "mode": "nvfp4",
39
+ "model.language_model.embed_tokens": {
40
+ "group_size": 64,
41
+ "bits": 4,
42
+ "mode": "affine"
43
+ },
44
+ "model.language_model.layers.0.self_attn.q_proj": {
45
+ "group_size": 64,
46
+ "bits": 4,
47
+ "mode": "affine"
48
+ },
49
+ "model.language_model.layers.0.self_attn.k_proj": {
50
+ "group_size": 64,
51
+ "bits": 4,
52
+ "mode": "affine"
53
+ },
54
+ "model.language_model.layers.0.self_attn.v_proj": {
55
+ "group_size": 64,
56
+ "bits": 4,
57
+ "mode": "affine"
58
+ },
59
+ "model.language_model.layers.0.self_attn.r_proj": {
60
+ "group_size": 64,
61
+ "bits": 4,
62
+ "mode": "affine"
63
+ },
64
+ "model.language_model.layers.0.self_attn.o_proj": {
65
+ "group_size": 64,
66
+ "bits": 4,
67
+ "mode": "affine"
68
+ },
69
+ "model.language_model.layers.0.self_attn.gate_proj": {
70
+ "group_size": 64,
71
+ "bits": 4,
72
+ "mode": "affine"
73
+ },
74
+ "model.language_model.layers.0.self_attn.sink_proj": {
75
+ "group_size": 64,
76
+ "bits": 4,
77
+ "mode": "affine"
78
+ },
79
+ "model.language_model.layers.1.self_attn.q_proj": {
80
+ "group_size": 64,
81
+ "bits": 4,
82
+ "mode": "affine"
83
+ },
84
+ "model.language_model.layers.1.self_attn.k_proj": {
85
+ "group_size": 64,
86
+ "bits": 4,
87
+ "mode": "affine"
88
+ },
89
+ "model.language_model.layers.1.self_attn.v_proj": {
90
+ "group_size": 64,
91
+ "bits": 4,
92
+ "mode": "affine"
93
+ },
94
+ "model.language_model.layers.1.self_attn.r_proj": {
95
+ "group_size": 64,
96
+ "bits": 4,
97
+ "mode": "affine"
98
+ },
99
+ "model.language_model.layers.1.self_attn.o_proj": {
100
+ "group_size": 64,
101
+ "bits": 4,
102
+ "mode": "affine"
103
+ },
104
+ "model.language_model.layers.1.self_attn.gate_proj": {
105
+ "group_size": 64,
106
+ "bits": 4,
107
+ "mode": "affine"
108
+ },
109
+ "model.language_model.layers.1.self_attn.sink_proj": {
110
+ "group_size": 64,
111
+ "bits": 4,
112
+ "mode": "affine"
113
+ },
114
+ "model.language_model.layers.2.self_attn.q_proj": {
115
+ "group_size": 64,
116
+ "bits": 4,
117
+ "mode": "affine"
118
+ },
119
+ "model.language_model.layers.2.self_attn.k_proj": {
120
+ "group_size": 64,
121
+ "bits": 4,
122
+ "mode": "affine"
123
+ },
124
+ "model.language_model.layers.2.self_attn.v_proj": {
125
+ "group_size": 64,
126
+ "bits": 4,
127
+ "mode": "affine"
128
+ },
129
+ "model.language_model.layers.2.self_attn.r_proj": {
130
+ "group_size": 64,
131
+ "bits": 4,
132
+ "mode": "affine"
133
+ },
134
+ "model.language_model.layers.2.self_attn.o_proj": {
135
+ "group_size": 64,
136
+ "bits": 4,
137
+ "mode": "affine"
138
+ },
139
+ "model.language_model.layers.2.self_attn.gate_proj": {
140
+ "group_size": 64,
141
+ "bits": 4,
142
+ "mode": "affine"
143
+ },
144
+ "model.language_model.layers.2.self_attn.sink_proj": {
145
+ "group_size": 64,
146
+ "bits": 4,
147
+ "mode": "affine"
148
+ },
149
+ "model.language_model.layers.3.self_attn.q_proj": {
150
+ "group_size": 64,
151
+ "bits": 4,
152
+ "mode": "affine"
153
+ },
154
+ "model.language_model.layers.3.self_attn.k_proj": {
155
+ "group_size": 64,
156
+ "bits": 4,
157
+ "mode": "affine"
158
+ },
159
+ "model.language_model.layers.3.self_attn.v_proj": {
160
+ "group_size": 64,
161
+ "bits": 4,
162
+ "mode": "affine"
163
+ },
164
+ "model.language_model.layers.3.self_attn.r_proj": {
165
+ "group_size": 64,
166
+ "bits": 4,
167
+ "mode": "affine"
168
+ },
169
+ "model.language_model.layers.3.self_attn.o_proj": {
170
+ "group_size": 64,
171
+ "bits": 4,
172
+ "mode": "affine"
173
+ },
174
+ "model.language_model.layers.3.self_attn.gate_proj": {
175
+ "group_size": 64,
176
+ "bits": 4,
177
+ "mode": "affine"
178
+ },
179
+ "model.language_model.layers.3.self_attn.sink_proj": {
180
+ "group_size": 64,
181
+ "bits": 4,
182
+ "mode": "affine"
183
+ },
184
+ "model.language_model.layers.4.self_attn.q_proj": {
185
+ "group_size": 64,
186
+ "bits": 4,
187
+ "mode": "affine"
188
+ },
189
+ "model.language_model.layers.4.self_attn.k_proj": {
190
+ "group_size": 64,
191
+ "bits": 4,
192
+ "mode": "affine"
193
+ },
194
+ "model.language_model.layers.4.self_attn.v_proj": {
195
+ "group_size": 64,
196
+ "bits": 4,
197
+ "mode": "affine"
198
+ },
199
+ "model.language_model.layers.4.self_attn.r_proj": {
200
+ "group_size": 64,
201
+ "bits": 4,
202
+ "mode": "affine"
203
+ },
204
+ "model.language_model.layers.4.self_attn.o_proj": {
205
+ "group_size": 64,
206
+ "bits": 4,
207
+ "mode": "affine"
208
+ },
209
+ "model.language_model.layers.4.self_attn.gate_proj": {
210
+ "group_size": 64,
211
+ "bits": 4,
212
+ "mode": "affine"
213
+ },
214
+ "model.language_model.layers.4.self_attn.sink_proj": {
215
+ "group_size": 64,
216
+ "bits": 4,
217
+ "mode": "affine"
218
+ },
219
+ "model.language_model.layers.5.self_attn.q_proj": {
220
+ "group_size": 64,
221
+ "bits": 4,
222
+ "mode": "affine"
223
+ },
224
+ "model.language_model.layers.5.self_attn.k_proj": {
225
+ "group_size": 64,
226
+ "bits": 4,
227
+ "mode": "affine"
228
+ },
229
+ "model.language_model.layers.5.self_attn.v_proj": {
230
+ "group_size": 64,
231
+ "bits": 4,
232
+ "mode": "affine"
233
+ },
234
+ "model.language_model.layers.5.self_attn.r_proj": {
235
+ "group_size": 64,
236
+ "bits": 4,
237
+ "mode": "affine"
238
+ },
239
+ "model.language_model.layers.5.self_attn.o_proj": {
240
+ "group_size": 64,
241
+ "bits": 4,
242
+ "mode": "affine"
243
+ },
244
+ "model.language_model.layers.5.self_attn.gate_proj": {
245
+ "group_size": 64,
246
+ "bits": 4,
247
+ "mode": "affine"
248
+ },
249
+ "model.language_model.layers.5.self_attn.sink_proj": {
250
+ "group_size": 64,
251
+ "bits": 4,
252
+ "mode": "affine"
253
+ },
254
+ "model.language_model.layers.5.self_attn.k_sconv.weight_proj": {
255
+ "group_size": 64,
256
+ "bits": 4,
257
+ "mode": "affine"
258
+ },
259
+ "model.language_model.layers.5.self_attn.v_sconv.weight_proj": {
260
+ "group_size": 64,
261
+ "bits": 4,
262
+ "mode": "affine"
263
+ },
264
+ "model.language_model.layers.6.self_attn.q_proj": {
265
+ "group_size": 64,
266
+ "bits": 4,
267
+ "mode": "affine"
268
+ },
269
+ "model.language_model.layers.6.self_attn.k_proj": {
270
+ "group_size": 64,
271
+ "bits": 4,
272
+ "mode": "affine"
273
+ },
274
+ "model.language_model.layers.6.self_attn.v_proj": {
275
+ "group_size": 64,
276
+ "bits": 4,
277
+ "mode": "affine"
278
+ },
279
+ "model.language_model.layers.6.self_attn.r_proj": {
280
+ "group_size": 64,
281
+ "bits": 4,
282
+ "mode": "affine"
283
+ },
284
+ "model.language_model.layers.6.self_attn.o_proj": {
285
+ "group_size": 64,
286
+ "bits": 4,
287
+ "mode": "affine"
288
+ },
289
+ "model.language_model.layers.6.self_attn.gate_proj": {
290
+ "group_size": 64,
291
+ "bits": 4,
292
+ "mode": "affine"
293
+ },
294
+ "model.language_model.layers.6.self_attn.sink_proj": {
295
+ "group_size": 64,
296
+ "bits": 4,
297
+ "mode": "affine"
298
+ },
299
+ "model.language_model.layers.7.self_attn.q_proj": {
300
+ "group_size": 64,
301
+ "bits": 4,
302
+ "mode": "affine"
303
+ },
304
+ "model.language_model.layers.7.self_attn.k_proj": {
305
+ "group_size": 64,
306
+ "bits": 4,
307
+ "mode": "affine"
308
+ },
309
+ "model.language_model.layers.7.self_attn.v_proj": {
310
+ "group_size": 64,
311
+ "bits": 4,
312
+ "mode": "affine"
313
+ },
314
+ "model.language_model.layers.7.self_attn.r_proj": {
315
+ "group_size": 64,
316
+ "bits": 4,
317
+ "mode": "affine"
318
+ },
319
+ "model.language_model.layers.7.self_attn.o_proj": {
320
+ "group_size": 64,
321
+ "bits": 4,
322
+ "mode": "affine"
323
+ },
324
+ "model.language_model.layers.7.self_attn.gate_proj": {
325
+ "group_size": 64,
326
+ "bits": 4,
327
+ "mode": "affine"
328
+ },
329
+ "model.language_model.layers.7.self_attn.sink_proj": {
330
+ "group_size": 64,
331
+ "bits": 4,
332
+ "mode": "affine"
333
+ },
334
+ "model.language_model.layers.8.self_attn.q_proj": {
335
+ "group_size": 64,
336
+ "bits": 4,
337
+ "mode": "affine"
338
+ },
339
+ "model.language_model.layers.8.self_attn.k_proj": {
340
+ "group_size": 64,
341
+ "bits": 4,
342
+ "mode": "affine"
343
+ },
344
+ "model.language_model.layers.8.self_attn.v_proj": {
345
+ "group_size": 64,
346
+ "bits": 4,
347
+ "mode": "affine"
348
+ },
349
+ "model.language_model.layers.8.self_attn.r_proj": {
350
+ "group_size": 64,
351
+ "bits": 4,
352
+ "mode": "affine"
353
+ },
354
+ "model.language_model.layers.8.self_attn.o_proj": {
355
+ "group_size": 64,
356
+ "bits": 4,
357
+ "mode": "affine"
358
+ },
359
+ "model.language_model.layers.8.self_attn.gate_proj": {
360
+ "group_size": 64,
361
+ "bits": 4,
362
+ "mode": "affine"
363
+ },
364
+ "model.language_model.layers.8.self_attn.sink_proj": {
365
+ "group_size": 64,
366
+ "bits": 4,
367
+ "mode": "affine"
368
+ },
369
+ "model.language_model.layers.9.self_attn.q_proj": {
370
+ "group_size": 64,
371
+ "bits": 4,
372
+ "mode": "affine"
373
+ },
374
+ "model.language_model.layers.9.self_attn.k_proj": {
375
+ "group_size": 64,
376
+ "bits": 4,
377
+ "mode": "affine"
378
+ },
379
+ "model.language_model.layers.9.self_attn.v_proj": {
380
+ "group_size": 64,
381
+ "bits": 4,
382
+ "mode": "affine"
383
+ },
384
+ "model.language_model.layers.9.self_attn.r_proj": {
385
+ "group_size": 64,
386
+ "bits": 4,
387
+ "mode": "affine"
388
+ },
389
+ "model.language_model.layers.9.self_attn.o_proj": {
390
+ "group_size": 64,
391
+ "bits": 4,
392
+ "mode": "affine"
393
+ },
394
+ "model.language_model.layers.9.self_attn.gate_proj": {
395
+ "group_size": 64,
396
+ "bits": 4,
397
+ "mode": "affine"
398
+ },
399
+ "model.language_model.layers.9.self_attn.sink_proj": {
400
+ "group_size": 64,
401
+ "bits": 4,
402
+ "mode": "affine"
403
+ },
404
+ "model.language_model.layers.10.self_attn.q_proj": {
405
+ "group_size": 64,
406
+ "bits": 4,
407
+ "mode": "affine"
408
+ },
409
+ "model.language_model.layers.10.self_attn.k_proj": {
410
+ "group_size": 64,
411
+ "bits": 4,
412
+ "mode": "affine"
413
+ },
414
+ "model.language_model.layers.10.self_attn.v_proj": {
415
+ "group_size": 64,
416
+ "bits": 4,
417
+ "mode": "affine"
418
+ },
419
+ "model.language_model.layers.10.self_attn.r_proj": {
420
+ "group_size": 64,
421
+ "bits": 4,
422
+ "mode": "affine"
423
+ },
424
+ "model.language_model.layers.10.self_attn.o_proj": {
425
+ "group_size": 64,
426
+ "bits": 4,
427
+ "mode": "affine"
428
+ },
429
+ "model.language_model.layers.10.self_attn.gate_proj": {
430
+ "group_size": 64,
431
+ "bits": 4,
432
+ "mode": "affine"
433
+ },
434
+ "model.language_model.layers.10.self_attn.sink_proj": {
435
+ "group_size": 64,
436
+ "bits": 4,
437
+ "mode": "affine"
438
+ },
439
+ "model.language_model.layers.11.self_attn.q_proj": {
440
+ "group_size": 64,
441
+ "bits": 4,
442
+ "mode": "affine"
443
+ },
444
+ "model.language_model.layers.11.self_attn.k_proj": {
445
+ "group_size": 64,
446
+ "bits": 4,
447
+ "mode": "affine"
448
+ },
449
+ "model.language_model.layers.11.self_attn.v_proj": {
450
+ "group_size": 64,
451
+ "bits": 4,
452
+ "mode": "affine"
453
+ },
454
+ "model.language_model.layers.11.self_attn.r_proj": {
455
+ "group_size": 64,
456
+ "bits": 4,
457
+ "mode": "affine"
458
+ },
459
+ "model.language_model.layers.11.self_attn.o_proj": {
460
+ "group_size": 64,
461
+ "bits": 4,
462
+ "mode": "affine"
463
+ },
464
+ "model.language_model.layers.11.self_attn.gate_proj": {
465
+ "group_size": 64,
466
+ "bits": 4,
467
+ "mode": "affine"
468
+ },
469
+ "model.language_model.layers.11.self_attn.sink_proj": {
470
+ "group_size": 64,
471
+ "bits": 4,
472
+ "mode": "affine"
473
+ },
474
+ "model.language_model.layers.11.self_attn.k_sconv.weight_proj": {
475
+ "group_size": 64,
476
+ "bits": 4,
477
+ "mode": "affine"
478
+ },
479
+ "model.language_model.layers.11.self_attn.v_sconv.weight_proj": {
480
+ "group_size": 64,
481
+ "bits": 4,
482
+ "mode": "affine"
483
+ },
484
+ "model.language_model.layers.12.self_attn.q_proj": {
485
+ "group_size": 64,
486
+ "bits": 4,
487
+ "mode": "affine"
488
+ },
489
+ "model.language_model.layers.12.self_attn.k_proj": {
490
+ "group_size": 64,
491
+ "bits": 4,
492
+ "mode": "affine"
493
+ },
494
+ "model.language_model.layers.12.self_attn.v_proj": {
495
+ "group_size": 64,
496
+ "bits": 4,
497
+ "mode": "affine"
498
+ },
499
+ "model.language_model.layers.12.self_attn.r_proj": {
500
+ "group_size": 64,
501
+ "bits": 4,
502
+ "mode": "affine"
503
+ },
504
+ "model.language_model.layers.12.self_attn.o_proj": {
505
+ "group_size": 64,
506
+ "bits": 4,
507
+ "mode": "affine"
508
+ },
509
+ "model.language_model.layers.12.self_attn.gate_proj": {
510
+ "group_size": 64,
511
+ "bits": 4,
512
+ "mode": "affine"
513
+ },
514
+ "model.language_model.layers.12.self_attn.sink_proj": {
515
+ "group_size": 64,
516
+ "bits": 4,
517
+ "mode": "affine"
518
+ },
519
+ "model.language_model.layers.13.self_attn.q_proj": {
520
+ "group_size": 64,
521
+ "bits": 4,
522
+ "mode": "affine"
523
+ },
524
+ "model.language_model.layers.13.self_attn.k_proj": {
525
+ "group_size": 64,
526
+ "bits": 4,
527
+ "mode": "affine"
528
+ },
529
+ "model.language_model.layers.13.self_attn.v_proj": {
530
+ "group_size": 64,
531
+ "bits": 4,
532
+ "mode": "affine"
533
+ },
534
+ "model.language_model.layers.13.self_attn.r_proj": {
535
+ "group_size": 64,
536
+ "bits": 4,
537
+ "mode": "affine"
538
+ },
539
+ "model.language_model.layers.13.self_attn.o_proj": {
540
+ "group_size": 64,
541
+ "bits": 4,
542
+ "mode": "affine"
543
+ },
544
+ "model.language_model.layers.13.self_attn.gate_proj": {
545
+ "group_size": 64,
546
+ "bits": 4,
547
+ "mode": "affine"
548
+ },
549
+ "model.language_model.layers.13.self_attn.sink_proj": {
550
+ "group_size": 64,
551
+ "bits": 4,
552
+ "mode": "affine"
553
+ },
554
+ "model.language_model.layers.14.self_attn.q_proj": {
555
+ "group_size": 64,
556
+ "bits": 4,
557
+ "mode": "affine"
558
+ },
559
+ "model.language_model.layers.14.self_attn.k_proj": {
560
+ "group_size": 64,
561
+ "bits": 4,
562
+ "mode": "affine"
563
+ },
564
+ "model.language_model.layers.14.self_attn.v_proj": {
565
+ "group_size": 64,
566
+ "bits": 4,
567
+ "mode": "affine"
568
+ },
569
+ "model.language_model.layers.14.self_attn.r_proj": {
570
+ "group_size": 64,
571
+ "bits": 4,
572
+ "mode": "affine"
573
+ },
574
+ "model.language_model.layers.14.self_attn.o_proj": {
575
+ "group_size": 64,
576
+ "bits": 4,
577
+ "mode": "affine"
578
+ },
579
+ "model.language_model.layers.14.self_attn.gate_proj": {
580
+ "group_size": 64,
581
+ "bits": 4,
582
+ "mode": "affine"
583
+ },
584
+ "model.language_model.layers.14.self_attn.sink_proj": {
585
+ "group_size": 64,
586
+ "bits": 4,
587
+ "mode": "affine"
588
+ },
589
+ "model.language_model.layers.15.self_attn.q_proj": {
590
+ "group_size": 64,
591
+ "bits": 4,
592
+ "mode": "affine"
593
+ },
594
+ "model.language_model.layers.15.self_attn.k_proj": {
595
+ "group_size": 64,
596
+ "bits": 4,
597
+ "mode": "affine"
598
+ },
599
+ "model.language_model.layers.15.self_attn.v_proj": {
600
+ "group_size": 64,
601
+ "bits": 4,
602
+ "mode": "affine"
603
+ },
604
+ "model.language_model.layers.15.self_attn.r_proj": {
605
+ "group_size": 64,
606
+ "bits": 4,
607
+ "mode": "affine"
608
+ },
609
+ "model.language_model.layers.15.self_attn.o_proj": {
610
+ "group_size": 64,
611
+ "bits": 4,
612
+ "mode": "affine"
613
+ },
614
+ "model.language_model.layers.15.self_attn.gate_proj": {
615
+ "group_size": 64,
616
+ "bits": 4,
617
+ "mode": "affine"
618
+ },
619
+ "model.language_model.layers.15.self_attn.sink_proj": {
620
+ "group_size": 64,
621
+ "bits": 4,
622
+ "mode": "affine"
623
+ },
624
+ "model.language_model.layers.16.self_attn.q_proj": {
625
+ "group_size": 64,
626
+ "bits": 4,
627
+ "mode": "affine"
628
+ },
629
+ "model.language_model.layers.16.self_attn.k_proj": {
630
+ "group_size": 64,
631
+ "bits": 4,
632
+ "mode": "affine"
633
+ },
634
+ "model.language_model.layers.16.self_attn.v_proj": {
635
+ "group_size": 64,
636
+ "bits": 4,
637
+ "mode": "affine"
638
+ },
639
+ "model.language_model.layers.16.self_attn.r_proj": {
640
+ "group_size": 64,
641
+ "bits": 4,
642
+ "mode": "affine"
643
+ },
644
+ "model.language_model.layers.16.self_attn.o_proj": {
645
+ "group_size": 64,
646
+ "bits": 4,
647
+ "mode": "affine"
648
+ },
649
+ "model.language_model.layers.16.self_attn.gate_proj": {
650
+ "group_size": 64,
651
+ "bits": 4,
652
+ "mode": "affine"
653
+ },
654
+ "model.language_model.layers.16.self_attn.sink_proj": {
655
+ "group_size": 64,
656
+ "bits": 4,
657
+ "mode": "affine"
658
+ },
659
+ "model.language_model.layers.17.self_attn.q_proj": {
660
+ "group_size": 64,
661
+ "bits": 4,
662
+ "mode": "affine"
663
+ },
664
+ "model.language_model.layers.17.self_attn.k_proj": {
665
+ "group_size": 64,
666
+ "bits": 4,
667
+ "mode": "affine"
668
+ },
669
+ "model.language_model.layers.17.self_attn.v_proj": {
670
+ "group_size": 64,
671
+ "bits": 4,
672
+ "mode": "affine"
673
+ },
674
+ "model.language_model.layers.17.self_attn.r_proj": {
675
+ "group_size": 64,
676
+ "bits": 4,
677
+ "mode": "affine"
678
+ },
679
+ "model.language_model.layers.17.self_attn.o_proj": {
680
+ "group_size": 64,
681
+ "bits": 4,
682
+ "mode": "affine"
683
+ },
684
+ "model.language_model.layers.17.self_attn.gate_proj": {
685
+ "group_size": 64,
686
+ "bits": 4,
687
+ "mode": "affine"
688
+ },
689
+ "model.language_model.layers.17.self_attn.sink_proj": {
690
+ "group_size": 64,
691
+ "bits": 4,
692
+ "mode": "affine"
693
+ },
694
+ "model.language_model.layers.17.self_attn.k_sconv.weight_proj": {
695
+ "group_size": 64,
696
+ "bits": 4,
697
+ "mode": "affine"
698
+ },
699
+ "model.language_model.layers.17.self_attn.v_sconv.weight_proj": {
700
+ "group_size": 64,
701
+ "bits": 4,
702
+ "mode": "affine"
703
+ },
704
+ "model.language_model.layers.18.self_attn.q_proj": {
705
+ "group_size": 64,
706
+ "bits": 4,
707
+ "mode": "affine"
708
+ },
709
+ "model.language_model.layers.18.self_attn.k_proj": {
710
+ "group_size": 64,
711
+ "bits": 4,
712
+ "mode": "affine"
713
+ },
714
+ "model.language_model.layers.18.self_attn.v_proj": {
715
+ "group_size": 64,
716
+ "bits": 4,
717
+ "mode": "affine"
718
+ },
719
+ "model.language_model.layers.18.self_attn.r_proj": {
720
+ "group_size": 64,
721
+ "bits": 4,
722
+ "mode": "affine"
723
+ },
724
+ "model.language_model.layers.18.self_attn.o_proj": {
725
+ "group_size": 64,
726
+ "bits": 4,
727
+ "mode": "affine"
728
+ },
729
+ "model.language_model.layers.18.self_attn.gate_proj": {
730
+ "group_size": 64,
731
+ "bits": 4,
732
+ "mode": "affine"
733
+ },
734
+ "model.language_model.layers.18.self_attn.sink_proj": {
735
+ "group_size": 64,
736
+ "bits": 4,
737
+ "mode": "affine"
738
+ },
739
+ "model.language_model.layers.19.self_attn.q_proj": {
740
+ "group_size": 64,
741
+ "bits": 4,
742
+ "mode": "affine"
743
+ },
744
+ "model.language_model.layers.19.self_attn.k_proj": {
745
+ "group_size": 64,
746
+ "bits": 4,
747
+ "mode": "affine"
748
+ },
749
+ "model.language_model.layers.19.self_attn.v_proj": {
750
+ "group_size": 64,
751
+ "bits": 4,
752
+ "mode": "affine"
753
+ },
754
+ "model.language_model.layers.19.self_attn.r_proj": {
755
+ "group_size": 64,
756
+ "bits": 4,
757
+ "mode": "affine"
758
+ },
759
+ "model.language_model.layers.19.self_attn.o_proj": {
760
+ "group_size": 64,
761
+ "bits": 4,
762
+ "mode": "affine"
763
+ },
764
+ "model.language_model.layers.19.self_attn.gate_proj": {
765
+ "group_size": 64,
766
+ "bits": 4,
767
+ "mode": "affine"
768
+ },
769
+ "model.language_model.layers.19.self_attn.sink_proj": {
770
+ "group_size": 64,
771
+ "bits": 4,
772
+ "mode": "affine"
773
+ },
774
+ "model.language_model.layers.20.self_attn.q_proj": {
775
+ "group_size": 64,
776
+ "bits": 4,
777
+ "mode": "affine"
778
+ },
779
+ "model.language_model.layers.20.self_attn.k_proj": {
780
+ "group_size": 64,
781
+ "bits": 4,
782
+ "mode": "affine"
783
+ },
784
+ "model.language_model.layers.20.self_attn.v_proj": {
785
+ "group_size": 64,
786
+ "bits": 4,
787
+ "mode": "affine"
788
+ },
789
+ "model.language_model.layers.20.self_attn.r_proj": {
790
+ "group_size": 64,
791
+ "bits": 4,
792
+ "mode": "affine"
793
+ },
794
+ "model.language_model.layers.20.self_attn.o_proj": {
795
+ "group_size": 64,
796
+ "bits": 4,
797
+ "mode": "affine"
798
+ },
799
+ "model.language_model.layers.20.self_attn.gate_proj": {
800
+ "group_size": 64,
801
+ "bits": 4,
802
+ "mode": "affine"
803
+ },
804
+ "model.language_model.layers.20.self_attn.sink_proj": {
805
+ "group_size": 64,
806
+ "bits": 4,
807
+ "mode": "affine"
808
+ },
809
+ "model.language_model.layers.21.self_attn.q_proj": {
810
+ "group_size": 64,
811
+ "bits": 4,
812
+ "mode": "affine"
813
+ },
814
+ "model.language_model.layers.21.self_attn.k_proj": {
815
+ "group_size": 64,
816
+ "bits": 4,
817
+ "mode": "affine"
818
+ },
819
+ "model.language_model.layers.21.self_attn.v_proj": {
820
+ "group_size": 64,
821
+ "bits": 4,
822
+ "mode": "affine"
823
+ },
824
+ "model.language_model.layers.21.self_attn.r_proj": {
825
+ "group_size": 64,
826
+ "bits": 4,
827
+ "mode": "affine"
828
+ },
829
+ "model.language_model.layers.21.self_attn.o_proj": {
830
+ "group_size": 64,
831
+ "bits": 4,
832
+ "mode": "affine"
833
+ },
834
+ "model.language_model.layers.21.self_attn.gate_proj": {
835
+ "group_size": 64,
836
+ "bits": 4,
837
+ "mode": "affine"
838
+ },
839
+ "model.language_model.layers.21.self_attn.sink_proj": {
840
+ "group_size": 64,
841
+ "bits": 4,
842
+ "mode": "affine"
843
+ },
844
+ "model.language_model.layers.22.self_attn.q_proj": {
845
+ "group_size": 64,
846
+ "bits": 4,
847
+ "mode": "affine"
848
+ },
849
+ "model.language_model.layers.22.self_attn.k_proj": {
850
+ "group_size": 64,
851
+ "bits": 4,
852
+ "mode": "affine"
853
+ },
854
+ "model.language_model.layers.22.self_attn.v_proj": {
855
+ "group_size": 64,
856
+ "bits": 4,
857
+ "mode": "affine"
858
+ },
859
+ "model.language_model.layers.22.self_attn.r_proj": {
860
+ "group_size": 64,
861
+ "bits": 4,
862
+ "mode": "affine"
863
+ },
864
+ "model.language_model.layers.22.self_attn.o_proj": {
865
+ "group_size": 64,
866
+ "bits": 4,
867
+ "mode": "affine"
868
+ },
869
+ "model.language_model.layers.22.self_attn.gate_proj": {
870
+ "group_size": 64,
871
+ "bits": 4,
872
+ "mode": "affine"
873
+ },
874
+ "model.language_model.layers.22.self_attn.sink_proj": {
875
+ "group_size": 64,
876
+ "bits": 4,
877
+ "mode": "affine"
878
+ },
879
+ "model.language_model.layers.23.self_attn.q_proj": {
880
+ "group_size": 64,
881
+ "bits": 4,
882
+ "mode": "affine"
883
+ },
884
+ "model.language_model.layers.23.self_attn.k_proj": {
885
+ "group_size": 64,
886
+ "bits": 4,
887
+ "mode": "affine"
888
+ },
889
+ "model.language_model.layers.23.self_attn.v_proj": {
890
+ "group_size": 64,
891
+ "bits": 4,
892
+ "mode": "affine"
893
+ },
894
+ "model.language_model.layers.23.self_attn.r_proj": {
895
+ "group_size": 64,
896
+ "bits": 4,
897
+ "mode": "affine"
898
+ },
899
+ "model.language_model.layers.23.self_attn.o_proj": {
900
+ "group_size": 64,
901
+ "bits": 4,
902
+ "mode": "affine"
903
+ },
904
+ "model.language_model.layers.23.self_attn.gate_proj": {
905
+ "group_size": 64,
906
+ "bits": 4,
907
+ "mode": "affine"
908
+ },
909
+ "model.language_model.layers.23.self_attn.sink_proj": {
910
+ "group_size": 64,
911
+ "bits": 4,
912
+ "mode": "affine"
913
+ },
914
+ "model.language_model.layers.23.self_attn.k_sconv.weight_proj": {
915
+ "group_size": 64,
916
+ "bits": 4,
917
+ "mode": "affine"
918
+ },
919
+ "model.language_model.layers.23.self_attn.v_sconv.weight_proj": {
920
+ "group_size": 64,
921
+ "bits": 4,
922
+ "mode": "affine"
923
+ },
924
+ "lm_head": {
925
+ "group_size": 64,
926
+ "bits": 4,
927
+ "mode": "affine"
928
+ }
929
+ },
930
+ "quantization_config": {
931
+ "group_size": 16,
932
+ "bits": 4,
933
+ "mode": "nvfp4",
934
+ "model.language_model.embed_tokens": {
935
+ "group_size": 64,
936
+ "bits": 4,
937
+ "mode": "affine"
938
+ },
939
+ "model.language_model.layers.0.self_attn.q_proj": {
940
+ "group_size": 64,
941
+ "bits": 4,
942
+ "mode": "affine"
943
+ },
944
+ "model.language_model.layers.0.self_attn.k_proj": {
945
+ "group_size": 64,
946
+ "bits": 4,
947
+ "mode": "affine"
948
+ },
949
+ "model.language_model.layers.0.self_attn.v_proj": {
950
+ "group_size": 64,
951
+ "bits": 4,
952
+ "mode": "affine"
953
+ },
954
+ "model.language_model.layers.0.self_attn.r_proj": {
955
+ "group_size": 64,
956
+ "bits": 4,
957
+ "mode": "affine"
958
+ },
959
+ "model.language_model.layers.0.self_attn.o_proj": {
960
+ "group_size": 64,
961
+ "bits": 4,
962
+ "mode": "affine"
963
+ },
964
+ "model.language_model.layers.0.self_attn.gate_proj": {
965
+ "group_size": 64,
966
+ "bits": 4,
967
+ "mode": "affine"
968
+ },
969
+ "model.language_model.layers.0.self_attn.sink_proj": {
970
+ "group_size": 64,
971
+ "bits": 4,
972
+ "mode": "affine"
973
+ },
974
+ "model.language_model.layers.1.self_attn.q_proj": {
975
+ "group_size": 64,
976
+ "bits": 4,
977
+ "mode": "affine"
978
+ },
979
+ "model.language_model.layers.1.self_attn.k_proj": {
980
+ "group_size": 64,
981
+ "bits": 4,
982
+ "mode": "affine"
983
+ },
984
+ "model.language_model.layers.1.self_attn.v_proj": {
985
+ "group_size": 64,
986
+ "bits": 4,
987
+ "mode": "affine"
988
+ },
989
+ "model.language_model.layers.1.self_attn.r_proj": {
990
+ "group_size": 64,
991
+ "bits": 4,
992
+ "mode": "affine"
993
+ },
994
+ "model.language_model.layers.1.self_attn.o_proj": {
995
+ "group_size": 64,
996
+ "bits": 4,
997
+ "mode": "affine"
998
+ },
999
+ "model.language_model.layers.1.self_attn.gate_proj": {
1000
+ "group_size": 64,
1001
+ "bits": 4,
1002
+ "mode": "affine"
1003
+ },
1004
+ "model.language_model.layers.1.self_attn.sink_proj": {
1005
+ "group_size": 64,
1006
+ "bits": 4,
1007
+ "mode": "affine"
1008
+ },
1009
+ "model.language_model.layers.2.self_attn.q_proj": {
1010
+ "group_size": 64,
1011
+ "bits": 4,
1012
+ "mode": "affine"
1013
+ },
1014
+ "model.language_model.layers.2.self_attn.k_proj": {
1015
+ "group_size": 64,
1016
+ "bits": 4,
1017
+ "mode": "affine"
1018
+ },
1019
+ "model.language_model.layers.2.self_attn.v_proj": {
1020
+ "group_size": 64,
1021
+ "bits": 4,
1022
+ "mode": "affine"
1023
+ },
1024
+ "model.language_model.layers.2.self_attn.r_proj": {
1025
+ "group_size": 64,
1026
+ "bits": 4,
1027
+ "mode": "affine"
1028
+ },
1029
+ "model.language_model.layers.2.self_attn.o_proj": {
1030
+ "group_size": 64,
1031
+ "bits": 4,
1032
+ "mode": "affine"
1033
+ },
1034
+ "model.language_model.layers.2.self_attn.gate_proj": {
1035
+ "group_size": 64,
1036
+ "bits": 4,
1037
+ "mode": "affine"
1038
+ },
1039
+ "model.language_model.layers.2.self_attn.sink_proj": {
1040
+ "group_size": 64,
1041
+ "bits": 4,
1042
+ "mode": "affine"
1043
+ },
1044
+ "model.language_model.layers.3.self_attn.q_proj": {
1045
+ "group_size": 64,
1046
+ "bits": 4,
1047
+ "mode": "affine"
1048
+ },
1049
+ "model.language_model.layers.3.self_attn.k_proj": {
1050
+ "group_size": 64,
1051
+ "bits": 4,
1052
+ "mode": "affine"
1053
+ },
1054
+ "model.language_model.layers.3.self_attn.v_proj": {
1055
+ "group_size": 64,
1056
+ "bits": 4,
1057
+ "mode": "affine"
1058
+ },
1059
+ "model.language_model.layers.3.self_attn.r_proj": {
1060
+ "group_size": 64,
1061
+ "bits": 4,
1062
+ "mode": "affine"
1063
+ },
1064
+ "model.language_model.layers.3.self_attn.o_proj": {
1065
+ "group_size": 64,
1066
+ "bits": 4,
1067
+ "mode": "affine"
1068
+ },
1069
+ "model.language_model.layers.3.self_attn.gate_proj": {
1070
+ "group_size": 64,
1071
+ "bits": 4,
1072
+ "mode": "affine"
1073
+ },
1074
+ "model.language_model.layers.3.self_attn.sink_proj": {
1075
+ "group_size": 64,
1076
+ "bits": 4,
1077
+ "mode": "affine"
1078
+ },
1079
+ "model.language_model.layers.4.self_attn.q_proj": {
1080
+ "group_size": 64,
1081
+ "bits": 4,
1082
+ "mode": "affine"
1083
+ },
1084
+ "model.language_model.layers.4.self_attn.k_proj": {
1085
+ "group_size": 64,
1086
+ "bits": 4,
1087
+ "mode": "affine"
1088
+ },
1089
+ "model.language_model.layers.4.self_attn.v_proj": {
1090
+ "group_size": 64,
1091
+ "bits": 4,
1092
+ "mode": "affine"
1093
+ },
1094
+ "model.language_model.layers.4.self_attn.r_proj": {
1095
+ "group_size": 64,
1096
+ "bits": 4,
1097
+ "mode": "affine"
1098
+ },
1099
+ "model.language_model.layers.4.self_attn.o_proj": {
1100
+ "group_size": 64,
1101
+ "bits": 4,
1102
+ "mode": "affine"
1103
+ },
1104
+ "model.language_model.layers.4.self_attn.gate_proj": {
1105
+ "group_size": 64,
1106
+ "bits": 4,
1107
+ "mode": "affine"
1108
+ },
1109
+ "model.language_model.layers.4.self_attn.sink_proj": {
1110
+ "group_size": 64,
1111
+ "bits": 4,
1112
+ "mode": "affine"
1113
+ },
1114
+ "model.language_model.layers.5.self_attn.q_proj": {
1115
+ "group_size": 64,
1116
+ "bits": 4,
1117
+ "mode": "affine"
1118
+ },
1119
+ "model.language_model.layers.5.self_attn.k_proj": {
1120
+ "group_size": 64,
1121
+ "bits": 4,
1122
+ "mode": "affine"
1123
+ },
1124
+ "model.language_model.layers.5.self_attn.v_proj": {
1125
+ "group_size": 64,
1126
+ "bits": 4,
1127
+ "mode": "affine"
1128
+ },
1129
+ "model.language_model.layers.5.self_attn.r_proj": {
1130
+ "group_size": 64,
1131
+ "bits": 4,
1132
+ "mode": "affine"
1133
+ },
1134
+ "model.language_model.layers.5.self_attn.o_proj": {
1135
+ "group_size": 64,
1136
+ "bits": 4,
1137
+ "mode": "affine"
1138
+ },
1139
+ "model.language_model.layers.5.self_attn.gate_proj": {
1140
+ "group_size": 64,
1141
+ "bits": 4,
1142
+ "mode": "affine"
1143
+ },
1144
+ "model.language_model.layers.5.self_attn.sink_proj": {
1145
+ "group_size": 64,
1146
+ "bits": 4,
1147
+ "mode": "affine"
1148
+ },
1149
+ "model.language_model.layers.5.self_attn.k_sconv.weight_proj": {
1150
+ "group_size": 64,
1151
+ "bits": 4,
1152
+ "mode": "affine"
1153
+ },
1154
+ "model.language_model.layers.5.self_attn.v_sconv.weight_proj": {
1155
+ "group_size": 64,
1156
+ "bits": 4,
1157
+ "mode": "affine"
1158
+ },
1159
+ "model.language_model.layers.6.self_attn.q_proj": {
1160
+ "group_size": 64,
1161
+ "bits": 4,
1162
+ "mode": "affine"
1163
+ },
1164
+ "model.language_model.layers.6.self_attn.k_proj": {
1165
+ "group_size": 64,
1166
+ "bits": 4,
1167
+ "mode": "affine"
1168
+ },
1169
+ "model.language_model.layers.6.self_attn.v_proj": {
1170
+ "group_size": 64,
1171
+ "bits": 4,
1172
+ "mode": "affine"
1173
+ },
1174
+ "model.language_model.layers.6.self_attn.r_proj": {
1175
+ "group_size": 64,
1176
+ "bits": 4,
1177
+ "mode": "affine"
1178
+ },
1179
+ "model.language_model.layers.6.self_attn.o_proj": {
1180
+ "group_size": 64,
1181
+ "bits": 4,
1182
+ "mode": "affine"
1183
+ },
1184
+ "model.language_model.layers.6.self_attn.gate_proj": {
1185
+ "group_size": 64,
1186
+ "bits": 4,
1187
+ "mode": "affine"
1188
+ },
1189
+ "model.language_model.layers.6.self_attn.sink_proj": {
1190
+ "group_size": 64,
1191
+ "bits": 4,
1192
+ "mode": "affine"
1193
+ },
1194
+ "model.language_model.layers.7.self_attn.q_proj": {
1195
+ "group_size": 64,
1196
+ "bits": 4,
1197
+ "mode": "affine"
1198
+ },
1199
+ "model.language_model.layers.7.self_attn.k_proj": {
1200
+ "group_size": 64,
1201
+ "bits": 4,
1202
+ "mode": "affine"
1203
+ },
1204
+ "model.language_model.layers.7.self_attn.v_proj": {
1205
+ "group_size": 64,
1206
+ "bits": 4,
1207
+ "mode": "affine"
1208
+ },
1209
+ "model.language_model.layers.7.self_attn.r_proj": {
1210
+ "group_size": 64,
1211
+ "bits": 4,
1212
+ "mode": "affine"
1213
+ },
1214
+ "model.language_model.layers.7.self_attn.o_proj": {
1215
+ "group_size": 64,
1216
+ "bits": 4,
1217
+ "mode": "affine"
1218
+ },
1219
+ "model.language_model.layers.7.self_attn.gate_proj": {
1220
+ "group_size": 64,
1221
+ "bits": 4,
1222
+ "mode": "affine"
1223
+ },
1224
+ "model.language_model.layers.7.self_attn.sink_proj": {
1225
+ "group_size": 64,
1226
+ "bits": 4,
1227
+ "mode": "affine"
1228
+ },
1229
+ "model.language_model.layers.8.self_attn.q_proj": {
1230
+ "group_size": 64,
1231
+ "bits": 4,
1232
+ "mode": "affine"
1233
+ },
1234
+ "model.language_model.layers.8.self_attn.k_proj": {
1235
+ "group_size": 64,
1236
+ "bits": 4,
1237
+ "mode": "affine"
1238
+ },
1239
+ "model.language_model.layers.8.self_attn.v_proj": {
1240
+ "group_size": 64,
1241
+ "bits": 4,
1242
+ "mode": "affine"
1243
+ },
1244
+ "model.language_model.layers.8.self_attn.r_proj": {
1245
+ "group_size": 64,
1246
+ "bits": 4,
1247
+ "mode": "affine"
1248
+ },
1249
+ "model.language_model.layers.8.self_attn.o_proj": {
1250
+ "group_size": 64,
1251
+ "bits": 4,
1252
+ "mode": "affine"
1253
+ },
1254
+ "model.language_model.layers.8.self_attn.gate_proj": {
1255
+ "group_size": 64,
1256
+ "bits": 4,
1257
+ "mode": "affine"
1258
+ },
1259
+ "model.language_model.layers.8.self_attn.sink_proj": {
1260
+ "group_size": 64,
1261
+ "bits": 4,
1262
+ "mode": "affine"
1263
+ },
1264
+ "model.language_model.layers.9.self_attn.q_proj": {
1265
+ "group_size": 64,
1266
+ "bits": 4,
1267
+ "mode": "affine"
1268
+ },
1269
+ "model.language_model.layers.9.self_attn.k_proj": {
1270
+ "group_size": 64,
1271
+ "bits": 4,
1272
+ "mode": "affine"
1273
+ },
1274
+ "model.language_model.layers.9.self_attn.v_proj": {
1275
+ "group_size": 64,
1276
+ "bits": 4,
1277
+ "mode": "affine"
1278
+ },
1279
+ "model.language_model.layers.9.self_attn.r_proj": {
1280
+ "group_size": 64,
1281
+ "bits": 4,
1282
+ "mode": "affine"
1283
+ },
1284
+ "model.language_model.layers.9.self_attn.o_proj": {
1285
+ "group_size": 64,
1286
+ "bits": 4,
1287
+ "mode": "affine"
1288
+ },
1289
+ "model.language_model.layers.9.self_attn.gate_proj": {
1290
+ "group_size": 64,
1291
+ "bits": 4,
1292
+ "mode": "affine"
1293
+ },
1294
+ "model.language_model.layers.9.self_attn.sink_proj": {
1295
+ "group_size": 64,
1296
+ "bits": 4,
1297
+ "mode": "affine"
1298
+ },
1299
+ "model.language_model.layers.10.self_attn.q_proj": {
1300
+ "group_size": 64,
1301
+ "bits": 4,
1302
+ "mode": "affine"
1303
+ },
1304
+ "model.language_model.layers.10.self_attn.k_proj": {
1305
+ "group_size": 64,
1306
+ "bits": 4,
1307
+ "mode": "affine"
1308
+ },
1309
+ "model.language_model.layers.10.self_attn.v_proj": {
1310
+ "group_size": 64,
1311
+ "bits": 4,
1312
+ "mode": "affine"
1313
+ },
1314
+ "model.language_model.layers.10.self_attn.r_proj": {
1315
+ "group_size": 64,
1316
+ "bits": 4,
1317
+ "mode": "affine"
1318
+ },
1319
+ "model.language_model.layers.10.self_attn.o_proj": {
1320
+ "group_size": 64,
1321
+ "bits": 4,
1322
+ "mode": "affine"
1323
+ },
1324
+ "model.language_model.layers.10.self_attn.gate_proj": {
1325
+ "group_size": 64,
1326
+ "bits": 4,
1327
+ "mode": "affine"
1328
+ },
1329
+ "model.language_model.layers.10.self_attn.sink_proj": {
1330
+ "group_size": 64,
1331
+ "bits": 4,
1332
+ "mode": "affine"
1333
+ },
1334
+ "model.language_model.layers.11.self_attn.q_proj": {
1335
+ "group_size": 64,
1336
+ "bits": 4,
1337
+ "mode": "affine"
1338
+ },
1339
+ "model.language_model.layers.11.self_attn.k_proj": {
1340
+ "group_size": 64,
1341
+ "bits": 4,
1342
+ "mode": "affine"
1343
+ },
1344
+ "model.language_model.layers.11.self_attn.v_proj": {
1345
+ "group_size": 64,
1346
+ "bits": 4,
1347
+ "mode": "affine"
1348
+ },
1349
+ "model.language_model.layers.11.self_attn.r_proj": {
1350
+ "group_size": 64,
1351
+ "bits": 4,
1352
+ "mode": "affine"
1353
+ },
1354
+ "model.language_model.layers.11.self_attn.o_proj": {
1355
+ "group_size": 64,
1356
+ "bits": 4,
1357
+ "mode": "affine"
1358
+ },
1359
+ "model.language_model.layers.11.self_attn.gate_proj": {
1360
+ "group_size": 64,
1361
+ "bits": 4,
1362
+ "mode": "affine"
1363
+ },
1364
+ "model.language_model.layers.11.self_attn.sink_proj": {
1365
+ "group_size": 64,
1366
+ "bits": 4,
1367
+ "mode": "affine"
1368
+ },
1369
+ "model.language_model.layers.11.self_attn.k_sconv.weight_proj": {
1370
+ "group_size": 64,
1371
+ "bits": 4,
1372
+ "mode": "affine"
1373
+ },
1374
+ "model.language_model.layers.11.self_attn.v_sconv.weight_proj": {
1375
+ "group_size": 64,
1376
+ "bits": 4,
1377
+ "mode": "affine"
1378
+ },
1379
+ "model.language_model.layers.12.self_attn.q_proj": {
1380
+ "group_size": 64,
1381
+ "bits": 4,
1382
+ "mode": "affine"
1383
+ },
1384
+ "model.language_model.layers.12.self_attn.k_proj": {
1385
+ "group_size": 64,
1386
+ "bits": 4,
1387
+ "mode": "affine"
1388
+ },
1389
+ "model.language_model.layers.12.self_attn.v_proj": {
1390
+ "group_size": 64,
1391
+ "bits": 4,
1392
+ "mode": "affine"
1393
+ },
1394
+ "model.language_model.layers.12.self_attn.r_proj": {
1395
+ "group_size": 64,
1396
+ "bits": 4,
1397
+ "mode": "affine"
1398
+ },
1399
+ "model.language_model.layers.12.self_attn.o_proj": {
1400
+ "group_size": 64,
1401
+ "bits": 4,
1402
+ "mode": "affine"
1403
+ },
1404
+ "model.language_model.layers.12.self_attn.gate_proj": {
1405
+ "group_size": 64,
1406
+ "bits": 4,
1407
+ "mode": "affine"
1408
+ },
1409
+ "model.language_model.layers.12.self_attn.sink_proj": {
1410
+ "group_size": 64,
1411
+ "bits": 4,
1412
+ "mode": "affine"
1413
+ },
1414
+ "model.language_model.layers.13.self_attn.q_proj": {
1415
+ "group_size": 64,
1416
+ "bits": 4,
1417
+ "mode": "affine"
1418
+ },
1419
+ "model.language_model.layers.13.self_attn.k_proj": {
1420
+ "group_size": 64,
1421
+ "bits": 4,
1422
+ "mode": "affine"
1423
+ },
1424
+ "model.language_model.layers.13.self_attn.v_proj": {
1425
+ "group_size": 64,
1426
+ "bits": 4,
1427
+ "mode": "affine"
1428
+ },
1429
+ "model.language_model.layers.13.self_attn.r_proj": {
1430
+ "group_size": 64,
1431
+ "bits": 4,
1432
+ "mode": "affine"
1433
+ },
1434
+ "model.language_model.layers.13.self_attn.o_proj": {
1435
+ "group_size": 64,
1436
+ "bits": 4,
1437
+ "mode": "affine"
1438
+ },
1439
+ "model.language_model.layers.13.self_attn.gate_proj": {
1440
+ "group_size": 64,
1441
+ "bits": 4,
1442
+ "mode": "affine"
1443
+ },
1444
+ "model.language_model.layers.13.self_attn.sink_proj": {
1445
+ "group_size": 64,
1446
+ "bits": 4,
1447
+ "mode": "affine"
1448
+ },
1449
+ "model.language_model.layers.14.self_attn.q_proj": {
1450
+ "group_size": 64,
1451
+ "bits": 4,
1452
+ "mode": "affine"
1453
+ },
1454
+ "model.language_model.layers.14.self_attn.k_proj": {
1455
+ "group_size": 64,
1456
+ "bits": 4,
1457
+ "mode": "affine"
1458
+ },
1459
+ "model.language_model.layers.14.self_attn.v_proj": {
1460
+ "group_size": 64,
1461
+ "bits": 4,
1462
+ "mode": "affine"
1463
+ },
1464
+ "model.language_model.layers.14.self_attn.r_proj": {
1465
+ "group_size": 64,
1466
+ "bits": 4,
1467
+ "mode": "affine"
1468
+ },
1469
+ "model.language_model.layers.14.self_attn.o_proj": {
1470
+ "group_size": 64,
1471
+ "bits": 4,
1472
+ "mode": "affine"
1473
+ },
1474
+ "model.language_model.layers.14.self_attn.gate_proj": {
1475
+ "group_size": 64,
1476
+ "bits": 4,
1477
+ "mode": "affine"
1478
+ },
1479
+ "model.language_model.layers.14.self_attn.sink_proj": {
1480
+ "group_size": 64,
1481
+ "bits": 4,
1482
+ "mode": "affine"
1483
+ },
1484
+ "model.language_model.layers.15.self_attn.q_proj": {
1485
+ "group_size": 64,
1486
+ "bits": 4,
1487
+ "mode": "affine"
1488
+ },
1489
+ "model.language_model.layers.15.self_attn.k_proj": {
1490
+ "group_size": 64,
1491
+ "bits": 4,
1492
+ "mode": "affine"
1493
+ },
1494
+ "model.language_model.layers.15.self_attn.v_proj": {
1495
+ "group_size": 64,
1496
+ "bits": 4,
1497
+ "mode": "affine"
1498
+ },
1499
+ "model.language_model.layers.15.self_attn.r_proj": {
1500
+ "group_size": 64,
1501
+ "bits": 4,
1502
+ "mode": "affine"
1503
+ },
1504
+ "model.language_model.layers.15.self_attn.o_proj": {
1505
+ "group_size": 64,
1506
+ "bits": 4,
1507
+ "mode": "affine"
1508
+ },
1509
+ "model.language_model.layers.15.self_attn.gate_proj": {
1510
+ "group_size": 64,
1511
+ "bits": 4,
1512
+ "mode": "affine"
1513
+ },
1514
+ "model.language_model.layers.15.self_attn.sink_proj": {
1515
+ "group_size": 64,
1516
+ "bits": 4,
1517
+ "mode": "affine"
1518
+ },
1519
+ "model.language_model.layers.16.self_attn.q_proj": {
1520
+ "group_size": 64,
1521
+ "bits": 4,
1522
+ "mode": "affine"
1523
+ },
1524
+ "model.language_model.layers.16.self_attn.k_proj": {
1525
+ "group_size": 64,
1526
+ "bits": 4,
1527
+ "mode": "affine"
1528
+ },
1529
+ "model.language_model.layers.16.self_attn.v_proj": {
1530
+ "group_size": 64,
1531
+ "bits": 4,
1532
+ "mode": "affine"
1533
+ },
1534
+ "model.language_model.layers.16.self_attn.r_proj": {
1535
+ "group_size": 64,
1536
+ "bits": 4,
1537
+ "mode": "affine"
1538
+ },
1539
+ "model.language_model.layers.16.self_attn.o_proj": {
1540
+ "group_size": 64,
1541
+ "bits": 4,
1542
+ "mode": "affine"
1543
+ },
1544
+ "model.language_model.layers.16.self_attn.gate_proj": {
1545
+ "group_size": 64,
1546
+ "bits": 4,
1547
+ "mode": "affine"
1548
+ },
1549
+ "model.language_model.layers.16.self_attn.sink_proj": {
1550
+ "group_size": 64,
1551
+ "bits": 4,
1552
+ "mode": "affine"
1553
+ },
1554
+ "model.language_model.layers.17.self_attn.q_proj": {
1555
+ "group_size": 64,
1556
+ "bits": 4,
1557
+ "mode": "affine"
1558
+ },
1559
+ "model.language_model.layers.17.self_attn.k_proj": {
1560
+ "group_size": 64,
1561
+ "bits": 4,
1562
+ "mode": "affine"
1563
+ },
1564
+ "model.language_model.layers.17.self_attn.v_proj": {
1565
+ "group_size": 64,
1566
+ "bits": 4,
1567
+ "mode": "affine"
1568
+ },
1569
+ "model.language_model.layers.17.self_attn.r_proj": {
1570
+ "group_size": 64,
1571
+ "bits": 4,
1572
+ "mode": "affine"
1573
+ },
1574
+ "model.language_model.layers.17.self_attn.o_proj": {
1575
+ "group_size": 64,
1576
+ "bits": 4,
1577
+ "mode": "affine"
1578
+ },
1579
+ "model.language_model.layers.17.self_attn.gate_proj": {
1580
+ "group_size": 64,
1581
+ "bits": 4,
1582
+ "mode": "affine"
1583
+ },
1584
+ "model.language_model.layers.17.self_attn.sink_proj": {
1585
+ "group_size": 64,
1586
+ "bits": 4,
1587
+ "mode": "affine"
1588
+ },
1589
+ "model.language_model.layers.17.self_attn.k_sconv.weight_proj": {
1590
+ "group_size": 64,
1591
+ "bits": 4,
1592
+ "mode": "affine"
1593
+ },
1594
+ "model.language_model.layers.17.self_attn.v_sconv.weight_proj": {
1595
+ "group_size": 64,
1596
+ "bits": 4,
1597
+ "mode": "affine"
1598
+ },
1599
+ "model.language_model.layers.18.self_attn.q_proj": {
1600
+ "group_size": 64,
1601
+ "bits": 4,
1602
+ "mode": "affine"
1603
+ },
1604
+ "model.language_model.layers.18.self_attn.k_proj": {
1605
+ "group_size": 64,
1606
+ "bits": 4,
1607
+ "mode": "affine"
1608
+ },
1609
+ "model.language_model.layers.18.self_attn.v_proj": {
1610
+ "group_size": 64,
1611
+ "bits": 4,
1612
+ "mode": "affine"
1613
+ },
1614
+ "model.language_model.layers.18.self_attn.r_proj": {
1615
+ "group_size": 64,
1616
+ "bits": 4,
1617
+ "mode": "affine"
1618
+ },
1619
+ "model.language_model.layers.18.self_attn.o_proj": {
1620
+ "group_size": 64,
1621
+ "bits": 4,
1622
+ "mode": "affine"
1623
+ },
1624
+ "model.language_model.layers.18.self_attn.gate_proj": {
1625
+ "group_size": 64,
1626
+ "bits": 4,
1627
+ "mode": "affine"
1628
+ },
1629
+ "model.language_model.layers.18.self_attn.sink_proj": {
1630
+ "group_size": 64,
1631
+ "bits": 4,
1632
+ "mode": "affine"
1633
+ },
1634
+ "model.language_model.layers.19.self_attn.q_proj": {
1635
+ "group_size": 64,
1636
+ "bits": 4,
1637
+ "mode": "affine"
1638
+ },
1639
+ "model.language_model.layers.19.self_attn.k_proj": {
1640
+ "group_size": 64,
1641
+ "bits": 4,
1642
+ "mode": "affine"
1643
+ },
1644
+ "model.language_model.layers.19.self_attn.v_proj": {
1645
+ "group_size": 64,
1646
+ "bits": 4,
1647
+ "mode": "affine"
1648
+ },
1649
+ "model.language_model.layers.19.self_attn.r_proj": {
1650
+ "group_size": 64,
1651
+ "bits": 4,
1652
+ "mode": "affine"
1653
+ },
1654
+ "model.language_model.layers.19.self_attn.o_proj": {
1655
+ "group_size": 64,
1656
+ "bits": 4,
1657
+ "mode": "affine"
1658
+ },
1659
+ "model.language_model.layers.19.self_attn.gate_proj": {
1660
+ "group_size": 64,
1661
+ "bits": 4,
1662
+ "mode": "affine"
1663
+ },
1664
+ "model.language_model.layers.19.self_attn.sink_proj": {
1665
+ "group_size": 64,
1666
+ "bits": 4,
1667
+ "mode": "affine"
1668
+ },
1669
+ "model.language_model.layers.20.self_attn.q_proj": {
1670
+ "group_size": 64,
1671
+ "bits": 4,
1672
+ "mode": "affine"
1673
+ },
1674
+ "model.language_model.layers.20.self_attn.k_proj": {
1675
+ "group_size": 64,
1676
+ "bits": 4,
1677
+ "mode": "affine"
1678
+ },
1679
+ "model.language_model.layers.20.self_attn.v_proj": {
1680
+ "group_size": 64,
1681
+ "bits": 4,
1682
+ "mode": "affine"
1683
+ },
1684
+ "model.language_model.layers.20.self_attn.r_proj": {
1685
+ "group_size": 64,
1686
+ "bits": 4,
1687
+ "mode": "affine"
1688
+ },
1689
+ "model.language_model.layers.20.self_attn.o_proj": {
1690
+ "group_size": 64,
1691
+ "bits": 4,
1692
+ "mode": "affine"
1693
+ },
1694
+ "model.language_model.layers.20.self_attn.gate_proj": {
1695
+ "group_size": 64,
1696
+ "bits": 4,
1697
+ "mode": "affine"
1698
+ },
1699
+ "model.language_model.layers.20.self_attn.sink_proj": {
1700
+ "group_size": 64,
1701
+ "bits": 4,
1702
+ "mode": "affine"
1703
+ },
1704
+ "model.language_model.layers.21.self_attn.q_proj": {
1705
+ "group_size": 64,
1706
+ "bits": 4,
1707
+ "mode": "affine"
1708
+ },
1709
+ "model.language_model.layers.21.self_attn.k_proj": {
1710
+ "group_size": 64,
1711
+ "bits": 4,
1712
+ "mode": "affine"
1713
+ },
1714
+ "model.language_model.layers.21.self_attn.v_proj": {
1715
+ "group_size": 64,
1716
+ "bits": 4,
1717
+ "mode": "affine"
1718
+ },
1719
+ "model.language_model.layers.21.self_attn.r_proj": {
1720
+ "group_size": 64,
1721
+ "bits": 4,
1722
+ "mode": "affine"
1723
+ },
1724
+ "model.language_model.layers.21.self_attn.o_proj": {
1725
+ "group_size": 64,
1726
+ "bits": 4,
1727
+ "mode": "affine"
1728
+ },
1729
+ "model.language_model.layers.21.self_attn.gate_proj": {
1730
+ "group_size": 64,
1731
+ "bits": 4,
1732
+ "mode": "affine"
1733
+ },
1734
+ "model.language_model.layers.21.self_attn.sink_proj": {
1735
+ "group_size": 64,
1736
+ "bits": 4,
1737
+ "mode": "affine"
1738
+ },
1739
+ "model.language_model.layers.22.self_attn.q_proj": {
1740
+ "group_size": 64,
1741
+ "bits": 4,
1742
+ "mode": "affine"
1743
+ },
1744
+ "model.language_model.layers.22.self_attn.k_proj": {
1745
+ "group_size": 64,
1746
+ "bits": 4,
1747
+ "mode": "affine"
1748
+ },
1749
+ "model.language_model.layers.22.self_attn.v_proj": {
1750
+ "group_size": 64,
1751
+ "bits": 4,
1752
+ "mode": "affine"
1753
+ },
1754
+ "model.language_model.layers.22.self_attn.r_proj": {
1755
+ "group_size": 64,
1756
+ "bits": 4,
1757
+ "mode": "affine"
1758
+ },
1759
+ "model.language_model.layers.22.self_attn.o_proj": {
1760
+ "group_size": 64,
1761
+ "bits": 4,
1762
+ "mode": "affine"
1763
+ },
1764
+ "model.language_model.layers.22.self_attn.gate_proj": {
1765
+ "group_size": 64,
1766
+ "bits": 4,
1767
+ "mode": "affine"
1768
+ },
1769
+ "model.language_model.layers.22.self_attn.sink_proj": {
1770
+ "group_size": 64,
1771
+ "bits": 4,
1772
+ "mode": "affine"
1773
+ },
1774
+ "model.language_model.layers.23.self_attn.q_proj": {
1775
+ "group_size": 64,
1776
+ "bits": 4,
1777
+ "mode": "affine"
1778
+ },
1779
+ "model.language_model.layers.23.self_attn.k_proj": {
1780
+ "group_size": 64,
1781
+ "bits": 4,
1782
+ "mode": "affine"
1783
+ },
1784
+ "model.language_model.layers.23.self_attn.v_proj": {
1785
+ "group_size": 64,
1786
+ "bits": 4,
1787
+ "mode": "affine"
1788
+ },
1789
+ "model.language_model.layers.23.self_attn.r_proj": {
1790
+ "group_size": 64,
1791
+ "bits": 4,
1792
+ "mode": "affine"
1793
+ },
1794
+ "model.language_model.layers.23.self_attn.o_proj": {
1795
+ "group_size": 64,
1796
+ "bits": 4,
1797
+ "mode": "affine"
1798
+ },
1799
+ "model.language_model.layers.23.self_attn.gate_proj": {
1800
+ "group_size": 64,
1801
+ "bits": 4,
1802
+ "mode": "affine"
1803
+ },
1804
+ "model.language_model.layers.23.self_attn.sink_proj": {
1805
+ "group_size": 64,
1806
+ "bits": 4,
1807
+ "mode": "affine"
1808
+ },
1809
+ "model.language_model.layers.23.self_attn.k_sconv.weight_proj": {
1810
+ "group_size": 64,
1811
+ "bits": 4,
1812
+ "mode": "affine"
1813
+ },
1814
+ "model.language_model.layers.23.self_attn.v_sconv.weight_proj": {
1815
+ "group_size": 64,
1816
+ "bits": 4,
1817
+ "mode": "affine"
1818
+ },
1819
+ "lm_head": {
1820
+ "group_size": 64,
1821
+ "bits": 4,
1822
+ "mode": "affine"
1823
+ }
1824
+ },
1825
+ "text_config": {
1826
+ "model_type": "inkling2_loop_text",
1827
+ "model_max_length": 1048576,
1828
+ "torch_dtype": "bfloat16",
1829
+ "hidden_size": 2304,
1830
+ "num_hidden_layers": 24,
1831
+ "vocab_size": 201024,
1832
+ "num_attention_heads": 24,
1833
+ "num_key_value_heads": 8,
1834
+ "head_dim": 128,
1835
+ "d_rel": 16,
1836
+ "rel_extent": 1024,
1837
+ "q_bias": false,
1838
+ "o_bias": false,
1839
+ "log_scaling_n_floor": 8192,
1840
+ "log_scaling_alpha": 0.1,
1841
+ "rms_norm_eps": 1e-06,
1842
+ "use_value_norm_no_gamma": true,
1843
+ "qk_l2_clip_threshold": 22.627416997969522,
1844
+ "rel_l2_clip_threshold": 8.0,
1845
+ "local_layer_ids": [
1846
+ 0,
1847
+ 1,
1848
+ 2,
1849
+ 3,
1850
+ 4,
1851
+ 6,
1852
+ 7,
1853
+ 8,
1854
+ 9,
1855
+ 10,
1856
+ 12,
1857
+ 13,
1858
+ 14,
1859
+ 15,
1860
+ 16,
1861
+ 18,
1862
+ 19,
1863
+ 20,
1864
+ 21,
1865
+ 22
1866
+ ],
1867
+ "sconv_kernel_size": 4,
1868
+ "unpadded_vocab_size": 200058,
1869
+ "final_logit_softcapping": 30.0,
1870
+ "swa_head_dim": 128,
1871
+ "swa_num_attention_heads": 24,
1872
+ "swa_num_key_value_heads": 8,
1873
+ "sliding_window_size": 1024,
1874
+ "n_layers_per_attn_res_block": 6,
1875
+ "attn_res_normalize_values": true,
1876
+ "n_routed_experts": 128,
1877
+ "num_experts_per_tok": 8,
1878
+ "n_shared_experts": 0,
1879
+ "intermediate_size": 1152,
1880
+ "mlp_gate_softcap": 4.0,
1881
+ "mlp_up_softcap": 25.0,
1882
+ "route_scale": 8.0,
1883
+ "use_gate_bias": true,
1884
+ "loop_config": {
1885
+ "core_start": 6,
1886
+ "core_end": 18,
1887
+ "num_passes": 3,
1888
+ "gate_rank": 64,
1889
+ "is_varloop": false
1890
+ }
1891
+ },
1892
+ "vision_config": {
1893
+ "model_type": "inkling2_vision",
1894
+ "text_hidden_size": 2304,
1895
+ "temporal_patch_size": 2,
1896
+ "native_patch_size": 20,
1897
+ "adapter_merge_size": 2,
1898
+ "projection_hidden_dim": 4096,
1899
+ "projector_bias": false,
1900
+ "hidden_size": 1024,
1901
+ "num_hidden_layers": 23,
1902
+ "num_attention_heads": 16,
1903
+ "mlp_ratio": 4.0,
1904
+ "posemb_grid_size": 32,
1905
+ "layer_norm_eps": 1e-05
1906
+ }
1907
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "temperature": 1.0,
4
+ "eos_token_id": 200006,
5
+ "pad_token_id": 199999
6
+ }
model-00001-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:acc233d29fdf2762fdc0edfb46e229be4a935611847e1105435f13a9ae7c0894
3
+ size 5333674631
model-00002-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:717b9ae80a043710944e3773a2ff76655f3b788d50ac88c1934bea1b05ee8c83
3
+ size 5031794004
model-00003-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f42cada3567f4613a971244f3bec5f1083d3ae3aca933512ae49643a6748139
3
+ size 4978917228
model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0aeeff012da96da5d5e6bf6f0b73a54de4c94bbc1c6321586cb4565469a02684
3
+ size 1963940606
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f332727f9a3f093dd452a39cc978e6757627e5dad1b0673aa4e75174f236dd6
3
+ size 26178678
tokenizer_config.json ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "begin_of_text": "<|begin_of_text|>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "content_image": "<|content_image|>",
6
+ "content_invoke_tool_json": "<|content_invoke_tool_json|>",
7
+ "content_invoke_tool_text": "<|content_invoke_tool_text|>",
8
+ "content_model_end_sampling": "<|content_model_end_sampling|>",
9
+ "content_text": "<|content_text|>",
10
+ "content_thinking": "<|content_thinking|>",
11
+ "content_tool_error": "<|content_tool_error|>",
12
+ "content_xml": "<|content_xml|>",
13
+ "end_message": "<|end_message|>",
14
+ "endoftext": "<|endoftext|>",
15
+ "eos_token": "<|content_model_end_sampling|>",
16
+ "fix_mistral_regex": false,
17
+ "is_local": true,
18
+ "local_files_only": false,
19
+ "message_model": "<|message_model|>",
20
+ "message_system": "<|message_system|>",
21
+ "message_tool": "<|message_tool|>",
22
+ "message_user": "<|message_user|>",
23
+ "model_max_length": 1048576,
24
+ "model_specific_special_tokens": {
25
+ "begin_of_text": "<|begin_of_text|>",
26
+ "content_image": "<|content_image|>",
27
+ "content_invoke_tool_json": "<|content_invoke_tool_json|>",
28
+ "content_invoke_tool_text": "<|content_invoke_tool_text|>",
29
+ "content_model_end_sampling": "<|content_model_end_sampling|>",
30
+ "content_text": "<|content_text|>",
31
+ "content_thinking": "<|content_thinking|>",
32
+ "content_tool_error": "<|content_tool_error|>",
33
+ "content_xml": "<|content_xml|>",
34
+ "end_message": "<|end_message|>",
35
+ "endoftext": "<|endoftext|>",
36
+ "message_model": "<|message_model|>",
37
+ "message_system": "<|message_system|>",
38
+ "message_tool": "<|message_tool|>",
39
+ "message_user": "<|message_user|>"
40
+ },
41
+ "pad_token": "<|endoftext|>",
42
+ "response_template": {
43
+ "defaults": {
44
+ "role": "assistant"
45
+ },
46
+ "fields": {
47
+ "content": {
48
+ "close": "<|end_message|>",
49
+ "content": "text",
50
+ "content_args": {
51
+ "strip": false
52
+ },
53
+ "join": "",
54
+ "open_pattern": "(?:<\\|message_model\\|>[^<]*)?<\\|content_(?:text|tool_error|invoke_tool_text)\\|>",
55
+ "repeats": true
56
+ },
57
+ "reasoning_content": {
58
+ "close": "<|end_message|>",
59
+ "content": "text",
60
+ "content_args": {
61
+ "strip": false
62
+ },
63
+ "join": "",
64
+ "open_pattern": "(?:<\\|message_model\\|>[^<]*)?<\\|content_thinking\\|>",
65
+ "repeats": true
66
+ },
67
+ "tool_calls": {
68
+ "close": "<|end_message|>",
69
+ "content": "json",
70
+ "open_pattern": "(?:<\\|message_model\\|>[^<]*)?<\\|content_invoke_tool_json\\|>",
71
+ "repeats": true,
72
+ "transform": {
73
+ "function": {
74
+ "arguments": "{content.args}",
75
+ "name": "{content.name}"
76
+ },
77
+ "type": "function"
78
+ }
79
+ }
80
+ },
81
+ "start_anchor_pattern": "<\\|end_message\\|>|<\\|content_model_end_sampling\\|>"
82
+ },
83
+ "tokenizer_class": "TokenizersBackend"
84
+ }