Add experimental quantized Josev-2B SystemOne GGUFs: Q8_0, Q4_K_MIX

.gitattributes

@@ -34,3 +34,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 josev2b-systemone-f32.gguf filter=lfs diff=lfs merge=lfs -text
+josev2b-systemone-q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+josev2b-systemone-q4_k-down8.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -15,3 +15,33 @@ arithmetic or quantization was performed. Upstream license and notices are kept
 - Not verified: task quality, other platforms, GPU, quantization. All seven model gates stay default-off.
 
 Modified from the upstream checkpoint by repackaging only (Apache-2.0 section 4).
+
+<!-- crispembed-experimental-quantized -->
+## Experimental quantized variants (owner decision 7 November 2026)
+
+**Experimental: not quality-qualified.** Use the F32 file above for exact typed responses. The
+quantized files were produced from the F32 file of revision `bffe62522f826b4c33f2e731d5b870ced5150a17` with
+CrispEmbed's C quantizer (tensor types changed; no weights trained or merged; Apache-2.0 section 4(b)).
+
+Quality vs the frozen original reference (four frozen typed requests, portable hosted CPU, receipt
+`tests/results/josev2b/q4k-sweep-41615751516.json`):
+
+| File | Precision | Bytes | Head-logit cosine (min) | Final-norm cosine (min) | Option argmax exact | Complete typed JSON exact | Max probability error | Strict 0.002/0.99999 stage budgets | Status |
+|---|---|---:|---:|---:|---:|---:|---:|---|---|
+| `josev2b-systemone-q8_0.gguf` | Q8_0 (`q8-all`; Q4_K/Q8_0/F32 tensors 0/150/292) | 3,630,289,376 | 0.99995 (bar > 0.97) | 0.99944 | 4/4 | 1/4 | 0.0059 | failed (200 stages) | experimental |
+| `josev2b-systemone-q4_k-down8.gguf` | Q4_K_MIX (`all-q4-down-q8`; Q4_K/Q8_0/F32 tensors 126/24/292) | 3,095,515,616 | 0.99739 (bar > 0.94) | 0.94911 | 4/4 | 0/4 | 0.0358 | failed (203 stages) | experimental |
+
+- Option choices match the original on 4/4 cases, but the rounded four-decimal probabilities and
+  scores move, so the complete typed JSON is not exact (Q8_0 1/4, Q4_K_MIX 0/4).
+- Not measured: independent task-quality panel, public-client (C/Python/CLI/HTTP/Rust/Dart) runs of
+  the quantized files, other platforms, GPU (the Josev runtime is CPU-only), latency/RSS.
+- Runtime: the same seven default-off Josev gates as F32 plus `CRISPEMBED_JOSEV_QUANT=1` and
+  `CRISPEMBED_QWEN35_QUANT=1`.
+
+- **Q8_0**: 150 projections Q8_0, 292 critical tensors F32; 3,462 MiB. Head-logit cosine min 0.99995, final-norm 0.99944 (four frozen cases, portable hosted CPU).
+- **Q4_K_MIX**: Named policy josev_q4k_down8_v1 (docs/josev2b-q4k-sweep.md): all decoder projections Q4_K except the 24 MLP down_proj at Q8_0; embedding, norms, head and other non-projection tensors F32. 2,952 MiB vs 3,462 MiB Q8_0. Head-logit cosine min 0.99739, final-norm 0.94911; option argmax 4/4, complete typed JSON 0/4 (rounded probabilities move).
+
+- `josev2b-systemone-q8_0.gguf`: SHA256 `fc1663bbd050938ff0bdb1b20ca52bd9ff7579dca63c0f7f9a15b3c3b7e36b85`
+- `josev2b-systemone-q4_k-down8.gguf`: SHA256 `adc7fa473febfef3fac10eac6fa0bebb6f68932675d1613e24706a3b74d5e582`
+
+Built by CrispEmbed commit `84b5fbedd8a1b07dd5e8a6582cc02010be75b11a` (workflow `josev-publish.yml`, protocol `docs/josev2b-q4k-publication-v1.json`).

josev2b-systemone-q4_k-down8.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adc7fa473febfef3fac10eac6fa0bebb6f68932675d1613e24706a3b74d5e582
+size 3095515616

josev2b-systemone-q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc1663bbd050938ff0bdb1b20ca52bd9ff7579dca63c0f7f9a15b3c3b7e36b85
+size 3630289376

manifest.json

@@ -18,5 +18,115 @@
     "tokenizer_config.json": "50470511708577d43f7c18c2d8e2ab02fb8abf3add1c40f60b74579d6d16e27e",
     "processor_config.json": "14932921ca485d458a04dafd8069fbb0a4505622a48208d19ed247115801385b"
   },
-  "consumer_receipt_sha256": "e8a2c4f2373d68fb87272c6d4fd0d8d302cf9ff3150dd722b83ede464bcdf2e5"
+  "consumer_receipt_sha256": "e8a2c4f2373d68fb87272c6d4fd0d8d302cf9ff3150dd722b83ede464bcdf2e5",
+  "experimental_quantized": {
+    "protocol": "docs/josev2b-q4k-publication-v1.json",
+    "protocol_sha256": "23580c1e4d5b72ed94e5f5525128da1a44bed16add07269eb8695d05f15a0dcd",
+    "source_commit": "84b5fbedd8a1b07dd5e8a6582cc02010be75b11a",
+    "from": {
+      "filename": "josev2b-systemone-f32.gguf",
+      "revision": "bffe62522f826b4c33f2e731d5b870ced5150a17",
+      "sha256": "72684e39d0eed526b40c5150c78b2d881d324c39e16af48060b3d741f526ccf2"
+    },
+    "files": [
+      {
+        "variant": "Q8_0",
+        "filename": "josev2b-systemone-q8_0.gguf",
+        "bytes": 3630289376,
+        "sha256": "fc1663bbd050938ff0bdb1b20ca52bd9ff7579dca63c0f7f9a15b3c3b7e36b85"
+      },
+      {
+        "variant": "Q4_K_MIX",
+        "filename": "josev2b-systemone-q4_k-down8.gguf",
+        "bytes": 3095515616,
+        "sha256": "adc7fa473febfef3fac10eac6fa0bebb6f68932675d1613e24706a3b74d5e582"
+      }
+    ],
+    "precision": {
+      "Q8_0": {
+        "variant": "Q8_0",
+        "recipe": "q8-all",
+        "receipt": "tests/results/josev2b/q4k-sweep-41615751516.json",
+        "receipt_sha256": "9dc81dbf9428a7156c8f7021d838cd79a76340f9a23c8fd9c29d16cecb80efb9",
+        "receipt_commit": "c2b72a58ef848bdcd4adf18bab6cd76a8284a3c8",
+        "bar": 0.97,
+        "head_logit_cos_min": 0.9999467750007115,
+        "final_norm_cos_min": 0.9994414401011729,
+        "final_norm_bar_passed": true,
+        "typed_json_exact": 1,
+        "argmax_exact": 4,
+        "cases": 4,
+        "probabilities_max_abs": 0.0058747828006744385,
+        "strict_stage_failures": 200,
+        "counts": {
+          "F32": 292,
+          "F16": 0,
+          "Q8_0": 150,
+          "Q4_K": 0
+        },
+        "expected_sha256": "fc1663bbd050938ff0bdb1b20ca52bd9ff7579dca63c0f7f9a15b3c3b7e36b85",
+        "expected_bytes": 3630289376
+      },
+      "Q4_K_MIX": {
+        "variant": "Q4_K_MIX",
+        "recipe": "all-q4-down-q8",
+        "receipt": "tests/results/josev2b/q4k-sweep-41615751516.json",
+        "receipt_sha256": "9dc81dbf9428a7156c8f7021d838cd79a76340f9a23c8fd9c29d16cecb80efb9",
+        "receipt_commit": "c2b72a58ef848bdcd4adf18bab6cd76a8284a3c8",
+        "bar": 0.94,
+        "head_logit_cos_min": 0.9973891795079161,
+        "final_norm_cos_min": 0.9491092304303442,
+        "final_norm_bar_passed": true,
+        "typed_json_exact": 0,
+        "argmax_exact": 4,
+        "cases": 4,
+        "probabilities_max_abs": 0.0358424186706543,
+        "strict_stage_failures": 203,
+        "counts": {
+          "F32": 292,
+          "F16": 0,
+          "Q8_0": 24,
+          "Q4_K": 126
+        },
+        "expected_sha256": "adc7fa473febfef3fac10eac6fa0bebb6f68932675d1613e24706a3b74d5e582",
+        "expected_bytes": 3095515616
+      }
+    },
+    "recipes": {
+      "Q8_0": {
+        "filename": "josev2b-systemone-q8_0.gguf",
+        "type": "q8_0",
+        "tensor_type": [
+          "*=f32",
+          "l.layers.*.self_attn.?_proj.weight=q8_0",
+          "l.layers.*.linear_attn.in_proj_qkv.weight=q8_0",
+          "l.layers.*.linear_attn.in_proj_z.weight=q8_0",
+          "l.layers.*.linear_attn.out_proj.weight=q8_0",
+          "l.layers.*.mlp.*_proj.weight=q8_0"
+        ],
+        "metadata_string": [
+          "josev2b.packed.format=crispembed-josev-packed-mixed-v2"
+        ],
+        "standard_file_type": true,
+        "note": "150 projections Q8_0, 292 critical tensors F32; 3,462 MiB. Head-logit cosine min 0.99995, final-norm 0.99944 (four frozen cases, portable hosted CPU).",
+        "precision_receipt": "tests/results/josev2b/q4k-sweep-41615751516.json",
+        "precision_dtype": "q8-all",
+        "precision_tier": "q8_0",
+        "precision_gate": "cosine"
+      },
+      "Q4_K_MIX": {
+        "filename": "josev2b-systemone-q4_k-down8.gguf",
+        "type": "q4_k",
+        "quantize_env": {
+          "CRISPEMBED_QUANTIZE_JOSEV_POLICY": "josev_q4k_down8_v1"
+        },
+        "note": "Named policy josev_q4k_down8_v1 (docs/josev2b-q4k-sweep.md): all decoder projections Q4_K except the 24 MLP down_proj at Q8_0; embedding, norms, head and other non-projection tensors F32. 2,952 MiB vs 3,462 MiB Q8_0. Head-logit cosine min 0.99739, final-norm 0.94911; option argmax 4/4, complete typed JSON 0/4 (rounded probabilities move).",
+        "precision_receipt": "tests/results/josev2b/q4k-sweep-41615751516.json",
+        "precision_dtype": "all-q4-down-q8",
+        "precision_tier": "q4_k",
+        "precision_gate": "cosine"
+      }
+    },
+    "previous_revision": "bffe62522f826b4c33f2e731d5b870ced5150a17"
+  }
 }