Duplicate from mizchi/laya-multilingual-onnx

Co-authored-by: Kotaro Chikuba <mizchi@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,47 @@
+---
+license: apache-2.0
+base_model: convaiinnovations/laya-multilingual
+library_name: onnxruntime
+tags:
+  - onnx
+  - onnxruntime-web
+  - webgpu
+  - typed-decisions
+  - laya
+---
+
+# Laya multilingual — ONNX (float16)
+
+ONNX export of [convaiinnovations/laya-multilingual](https://huggingface.co/convaiinnovations/laya-multilingual)
+(mmBERT-base encoder, 322M parameters) for onnxruntime, including onnxruntime-web with the WebGPU backend.
+Produced by [laya-mlx](https://github.com/mizorewww/laya-mlx) with `laya-mlx export-onnx --dtype float16` from the
+published [aac6fef/laya-multilingual-mlx](https://huggingface.co/aac6fef/laya-multilingual-mlx) checkpoint
+(snapshot `f2b4faf51023039425946074e2cf1361d2db11d5`, itself converted from upstream revision
+`052592a15d198d9ad47da779604259b10b47b7aa`).
+
+| File | Purpose |
+|---|---|
+| `model.onnx` | Encoder, decision head, scorer and action head in one graph. Opset 18, standard operators only, float16 weights with a float32 decision tail. |
+| `rl_agent_config.json` | Prompt limits (`max_len`, `head_max_len`) and calibration temperatures. |
+| `tokenizer/` | Hugging Face `tokenizer.json` and `tokenizer_config.json`. |
+| `onnx_config.json` | Export metadata: dtype, opset, input and output names, source checkpoint. |
+
+Inputs: `input_ids`, `attention_mask` (int64 `[batch, sequence]`), `marker_pos` (int64 `[batch, markers]`),
+`marker_mask` (bool `[batch, markers]`), `qtype` (int64 `[batch]`; 0 choice, 1 score, 2 noul). Outputs:
+`logits` `[batch, markers]` and `act_logits` `[batch, actions]`, both float32. `logits` still need the
+calibration temperature and a softmax. Prompt construction, calibration and answer formatting follow upstream
+Laya; the reference implementation is `laya_mlx.common` / `laya_mlx.agent` and the browser port is
+`@laya-mlx/web` in the repository.
+
+Validation on the repository's 63-question parity fixtures against the MLX runtime: 63/63 selected answers
+agree, maximum probability error 5.1e-4 with the onnxruntime CPU provider and 1.2e-2 with onnxruntime-web
+WebGPU (float16 compute). See `docs/ONNX_EXPORT.md` in the repository for the method, the browser
+measurements and the size analysis. onnxruntime-web 1.30.0 needs `graphOptimizationLevel: "basic"` on
+WebGPU for this graph.
+
+This is an independent port, not an official Convai Innovations release. Weights are Apache-2.0 as upstream.
+
+Live demo: https://huggingface.co/spaces/mizchi/laya-web-demo (Snake and Chess, both run the model in your
+browser — direct pages: https://mizchi-laya-web-demo.static.hf.space/snake.html and
+https://mizchi-laya-web-demo.static.hf.space/chess.html) — also on the repository's GitHub Pages site (for the
+`mizchi/laya-mlx` fork: https://mizchi.github.io/laya-mlx/).

model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b095e005a4c295cae74d47b7eb6931c369d48f5b720b45278d774165798310c
+size 646870871

onnx_config.json

@@ -0,0 +1,20 @@
+{
+  "format": "laya-onnx",
+  "format_version": 1,
+  "dtype": "float16",
+  "opset": 18,
+  "inputs": [
+    "input_ids",
+    "attention_mask",
+    "marker_pos",
+    "marker_mask",
+    "qtype"
+  ],
+  "outputs": [
+    "logits",
+    "act_logits"
+  ],
+  "source": "aac6fef/laya-multilingual-mlx",
+  "revision": null,
+  "subfolder": null
+}

rl_agent_config.json

@@ -0,0 +1,26 @@
+{
+  "encoder": "jhu-clsp/mmBERT-base",
+  "head_layers": 2,
+  "max_len": 1024,
+  "head_max_len": 256,
+  "max_prefixes": 6,
+  "act_costs": {
+    "escalate": 0.5
+  },
+  "cost_wrong_act": 3.0,
+  "amp_dtype": "bf16",
+  "model_name": "rl-agent",
+  "temperature": [
+    1.0,
+    1.0,
+    1.0
+  ],
+  "temperature_by_options": {},
+  "training": {
+    "updates": 15987,
+    "epochs_completed": 4,
+    "hours": 4.97,
+    "world_size": 1,
+    "fine_tuned_from_checkpoint": false
+  }
+}

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f
+size 34363188

tokenizer/tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<bos>",
+  "eos_token": "<eos>",
+  "extra_special_tokens": {
+    "extra_0": "<start_of_turn>",
+    "extra_1": "<end_of_turn>"
+  },
+  "mask_token": "<mask>",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "sep_token": "<eos>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>"
+}