feat: add ONNX quantized variants (8 models, 21MB–138MB)

- full FP32 (138 MB): best accuracy, no quantization loss
- fp16 (69 MB): half precision, minimal accuracy loss
- int8 / uint8 / quantized (36 MB): balanced speed/accuracy, best for CPU
- q4 (24 MB): aggressive 4-bit quantization
- bnb4 (22 MB): bitsandbytes 4-bit
- q4f16 (21 MB): smallest, 4-bit with fp16 compute

onnx/model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7540dd30b019c4a5dffab9447598f2bdd2316a3d5b3230cf90618b9c9475130
+size 143786825

onnx/model_bnb4.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f7e39e7859913de13d2f9625720c5e180dba7700369dff9dfb451fc19dfc362
+size 22146828

onnx/model_fp16.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2a67eb38414b0ea4b7c3dbaf4f23eda94ab4a6159ed378359a9ad1be1fb75e2
+size 71972090

onnx/model_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97b418289121c7644199f0bd35de9d0c3d319289393e7ba63ba01cefe1e5127c
+size 37124981

onnx/model_q4.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:693a2ff2e1b788ca01949bdf880d112c4aa7758cd1335fd7daa41c3f74ed1f99
+size 24358188

onnx/model_q4f16.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:602788a04a1de0d8ee7732219f97cece14a74f8a3bb5ca00207e38d70ece765d
+size 21110397

onnx/model_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09783d3e3e394838690eb8626741c7af5d72a1e3e9b88894ad01b007e94e14b9
+size 37124981

onnx/model_uint8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09783d3e3e394838690eb8626741c7af5d72a1e3e9b88894ad01b007e94e14b9
+size 37124981

onnx/quantize_config.json

@@ -0,0 +1,18 @@
+{
+    "modes": [
+        "fp16",
+        "q8",
+        "int8",
+        "uint8",
+        "q4",
+        "q4f16",
+        "bnb4"
+    ],
+    "per_channel": true,
+    "reduce_range": true,
+    "block_size": null,
+    "is_symmetric": true,
+    "accuracy_level": null,
+    "quant_type": 1,
+    "op_block_list": null
+}