raw float weights: model.safetensors + config.json, exported from the shipped model.onnx

README.md

@@ -467,9 +467,15 @@ codesearchnet python 68.30 → 68.32.
 - **`model.onnx`** — the FP32 twin of the same weights, for lanes that cannot read Q/DQ and for
   fine-tuning. Derived from the file above by dropping its Q/DQ pairs, so the weights are
   identical — the 0.98 cosine between them is the quantization, not a different model.
+- **`model.safetensors`** + **`config.json`** — the same float weights under transformers names,
+  exported from `model.onnx` and checked against it (max |Δ| 3e-7 on the pooled vector).
+  `AutoModel` loads the encoder (`XLMRobertaModel`); the trained projection `proj.weight` is
+  not part of that class, so apply it yourself: mean-pool over the attention mask → `proj` →
+  L2. `config.json` → `ultracode` spells the head out. This pair is also what the Apple (MLX)
+  build is prepared from.
 - **`sentencepiece.bpe.model`** — the tokenizer. Raw-SP ids: pad=0, unk=1, bos=2, eos=3. Its
   22,739 pieces match the embedding table row for row.
-- **`tokenizer_config.json`**, **`config.json`** — HF-side metadata.
+- **`tokenizer_config.json`** — HF-side metadata.
 
 **Prebuilt engines**, uploaded progressively per architecture:
 
@@ -478,7 +484,7 @@ TensorRT    …-{s,m,l,xl}_{win_x64,linux_x64}_trt11.0_sm_{75,80,86,89,120}.engi
             …-{s,m,l,xl}_linux_x64_trt11.0_sm_90.engine
 OpenVINO    …-{s,m,l,xl}_ov2026.4_{cpu_int8,igpu_lnl_int8,npu_int4}_b*_s*.{xml,bin}
 TVM Vulkan  …_{win_x64,linux_x64}_tvm0.25_vulkan.{dll,so}
-MLX (Apple) model_gpu_mlx0.22/
+MLX (Apple) model_gpu_mlx0.32/
 ```
 
 `sm_90` (H100 / H200) is Linux-only. An INT4 IR for discrete Intel Arc (`igpu_arc_int4`) is

config.json

@@ -20,5 +20,18 @@
   "transformers_version": "5.1.0",
   "type_vocab_size": 1,
   "use_cache": true,
-  "vocab_size": 22739
+  "vocab_size": 22739,
+  "architectures": [
+    "XLMRobertaModel"
+  ],
+  "ultracode": {
+    "format": "raw-safetensors/1",
+    "exported_from": "model.onnx",
+    "onnx_sha256": "57bcfc6aed11ea239d01f2b124f2f948456f2284ad6e2c4744452509c9c25ca9",
+    "head": "mean_proj_l2",
+    "position_ids": "roberta",
+    "pooling": "mean over attention_mask",
+    "projection": "proj.weight, applied after pooling",
+    "normalize": "l2"
+  }
 }

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:704f15d3ddddac9cfbc0b5c84ec387ef36593dd24c466f4184452b15d5253dc0
+size 187215600