Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* 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
+variants/cpu-int4/inference/frozen_base.onnx.data filter=lfs diff=lfs merge=lfs -text
+variants/cpu-int4/train/checkpoint filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,49 @@
+---
+base_model: HuggingFaceTB/SmolLM2-135M-Instruct
+library_name: mobiletransformers
+pipeline_tag: text-generation
+tags:
+  - mobiletransformers
+  - onnx
+  - on-device
+  - android
+  - lora
+  - int4
+---
+
+# HuggingFaceTB/SmolLM2-135M-Instruct — MobileTransformers package
+
+On-device (Android) package exported from **HuggingFaceTB/SmolLM2-135M-Instruct** with MobileTransformers.
+
+## What this package can do
+- `core` — shared files every other group needs
+- `inference` — generate or score on device
+- `train` — **fine-tune on device**, then merge the adapter back into the base weights
+- `rag` — retrieve over documents you ingest, and ground answers in them
+
+## Fine-tuning method
+- **lora** — LoRA — low-rank adapters on the attention projections.
+- Rank: `8`
+- Adapted modules: `q_proj`, `v_proj`
+
+## Provenance
+- Base model: `HuggingFaceTB/SmolLM2-135M-Instruct`
+- Selected task: `text-generation-with-past`
+- Quantization: int4
+- Toolchain: optimum-onnx 0.1.0, transformers 4.57.6, ort-training 1.23.0+cpu
+
+## Licenses
+- Framework: not declared in this package — see the repository
+- Base model weights: see the base model above (this package redistributes an export of those weights, so their terms govern its contents)
+
+## Android runtime
+- Minimum API: 28
+- Required ABIs: any
+
+## Variants
+
+| id | EP | quant | engines | features | min API | rec. RAM (MB) |
+| --- | --- | --- | --- | --- | --- | --- |
+| cpu-int4 | cpu | int4 | native | core, inference, train, rag | 28 | — |
+
+Default variant: `cpu-int4`.

mobiletransformers_manifest.json

@@ -0,0 +1,409 @@
+{
+  "androidRuntime": {
+    "minimumAndroidApi": 28,
+    "recommendedDeviceMemoryMb": null,
+    "requiredAbis": []
+  },
+  "architectures": [
+    "llama"
+  ],
+  "artifactFormatVersion": 1,
+  "baseModelId": "HuggingFaceTB/SmolLM2-135M-Instruct",
+  "defaultVariant": "cpu-int4",
+  "downloadPlan": {
+    "cpu-int4": {
+      "checksums": [
+        "variants/cpu-int4/checksums.json"
+      ],
+      "core": [
+        "mobiletransformers_manifest.json",
+        "shared/tokenizer/**",
+        "shared/chat_template.jinja",
+        "shared/config.json",
+        "shared/generation_config.json"
+      ],
+      "genai": [],
+      "inference": [
+        "variants/cpu-int4/inference/**"
+      ],
+      "rag": [
+        "variants/cpu-int4/embedding/**"
+      ],
+      "train": [
+        "variants/cpu-int4/train/**"
+      ]
+    }
+  },
+  "exportedAt": "",
+  "fileSizes": {
+    "optimum/export_report.json": 816,
+    "optimum/supported_tasks.json": 34,
+    "shared/tokenizer/chat_template.jinja": 368,
+    "shared/tokenizer/merges.txt": 466391,
+    "shared/tokenizer/mobiletransformers_tokenizer_config.json": 291,
+    "shared/tokenizer/special_tokens_map.json": 655,
+    "shared/tokenizer/tokenizer.json": 3522656,
+    "shared/tokenizer/tokenizer_config.json": 3398,
+    "shared/tokenizer/vocab.json": 800662,
+    "variants/cpu-int4/checksums.json": 21734,
+    "variants/cpu-int4/embedding/embedding_model.onnx": 90392834,
+    "variants/cpu-int4/embedding/rag_config.json": 343,
+    "variants/cpu-int4/embedding/tokenizer/special_tokens_map.json": 695,
+    "variants/cpu-int4/embedding/tokenizer/tokenizer.json": 711661,
+    "variants/cpu-int4/embedding/tokenizer/tokenizer_config.json": 1464,
+    "variants/cpu-int4/inference/chat_template.jinja": 368,
+    "variants/cpu-int4/inference/config.json": 872,
+    "variants/cpu-int4/inference/frozen_base.onnx.data": 598222080,
+    "variants/cpu-int4/inference/genai_config.json": 1010,
+    "variants/cpu-int4/inference/generation_config.json": 132,
+    "variants/cpu-int4/inference/merger_lora_fpin_fpout.onnx": 811,
+    "variants/cpu-int4/inference/merges.txt": 466391,
+    "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin": 1327104,
+    "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin": 442368,
+    "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin.sha256": 65,
+    "variants/cpu-int4/inference/model.onnx": 1257698,
+    "variants/cpu-int4/inference/optimum_config.json": 335,
+    "variants/cpu-int4/inference/special_tokens_map.json": 655,
+    "variants/cpu-int4/inference/tokenizer.json": 3522656,
+    "variants/cpu-int4/inference/tokenizer_config.json": 3398,
+    "variants/cpu-int4/inference/vocab.json": 800662,
+    "variants/cpu-int4/inference/weight_handoff_map.json": 89523,
+    "variants/cpu-int4/train/checkpoint": 176349432,
+    "variants/cpu-int4/train/eval_model.onnx": 2541985,
+    "variants/cpu-int4/train/optimizer_model.onnx": 538,
+    "variants/cpu-int4/train/trainable_parameters.json": 11643,
+    "variants/cpu-int4/train/training_config.json": 32999,
+    "variants/cpu-int4/train/training_model.onnx": 2659395
+  },
+  "license": {
+    "baseModelWeights": null,
+    "framework": null,
+    "noticeFile": null
+  },
+  "minReaderVersion": "1.0",
+  "mobiletransformersVersion": "0.1.0",
+  "onnxRuntimeGenAIVersion": null,
+  "onnxRuntimeTrainingVersion": "1.23.0+cpu",
+  "optimumOnnxVersion": "0.1.0",
+  "peftMethods": [
+    "lora"
+  ],
+  "quantization": [
+    "int4"
+  ],
+  "requiredFiles": [
+    "mobiletransformers_manifest.json",
+    "shared/tokenizer/tokenizer.json",
+    "variants/cpu-int4/inference/model.onnx"
+  ],
+  "schemaVersion": "1.0",
+  "selectedTask": "text-generation-with-past",
+  "sha256": {
+    "optimum/export_report.json": "a1fcd95a9305415a0bcc013e2a85335a40fb5463e769540e2e4bab6f459d9d5c",
+    "optimum/supported_tasks.json": "2591d52d3133d5bd71e384d09e57f48d29e84f4e9a358697595bb4bcb56dcfdf",
+    "shared/tokenizer/chat_template.jinja": "872be49dbb638044ad01b60388f48d469ff2980e5f0dccdc22ec907db54d0788",
+    "shared/tokenizer/merges.txt": "0b54e8aa4e53d5383e2e4bc635a56b43f9647f7b13832d5d9ecd8f82dac4f510",
+    "shared/tokenizer/mobiletransformers_tokenizer_config.json": "41a90dab627f46eb5c67a3e18134d224bb54e21dbfd6e4f9885a4b3187f7c638",
+    "shared/tokenizer/special_tokens_map.json": "2b7379f3ae813529281a5c602bc5a11c1d4e0a99107aaa597fe936c1e813ca52",
+    "shared/tokenizer/tokenizer.json": "7d27c493c729a66ecefc837280b05d948b1ed50d130eebdbf911b1b36cf38ed7",
+    "shared/tokenizer/tokenizer_config.json": "5dbff37e2d5899f5f69d86116f80733e11a790f703fd59855e2aef5717533ac5",
+    "shared/tokenizer/vocab.json": "82b84012e3add4d01d12ba14442026e49b8cbbaead1f79ecf3d919784f82dc79",
+    "variants/cpu-int4/checksums.json": "3ccab69b2c70d4e4256423700efed901a4dbf38a31f68f8b47777fd9cdf3fbd9",
+    "variants/cpu-int4/embedding/embedding_model.onnx": "7c46645e1dcec3d3a27d80f21fcad81ab307317bf3d0580eb8ea6779fa83cc96",
+    "variants/cpu-int4/embedding/rag_config.json": "f90d993929e9e7b44c0cbe4518c72e7c64979a527a9d91a766dc967dea95de16",
+    "variants/cpu-int4/embedding/tokenizer/special_tokens_map.json": "5d5b662e421ea9fac075174bb0688ee0d9431699900b90662acd44b2a350503a",
+    "variants/cpu-int4/embedding/tokenizer/tokenizer.json": "da0e79933b9ed51798a3ae27893d3c5fa4a201126cef75586296df9b4d2c62a0",
+    "variants/cpu-int4/embedding/tokenizer/tokenizer_config.json": "ccb4eb21a03e1442ee5c3f85431b9c307960a04579942537d74778dc8080a48c",
+    "variants/cpu-int4/inference/chat_template.jinja": "872be49dbb638044ad01b60388f48d469ff2980e5f0dccdc22ec907db54d0788",
+    "variants/cpu-int4/inference/config.json": "2f679c28417584f0b3a2d79542dd7da73f27cf927889434afa6b1731a091ef8e",
+    "variants/cpu-int4/inference/frozen_base.onnx.data": "12f4d16fce65a2186843b4b435bced33506c16b0f9a43dc1fad5388d99576548",
+    "variants/cpu-int4/inference/genai_config.json": "4a80b27a1e0bcde8917b7b15c2234813f48e1b894b7768f7fa68e07ff7e558d9",
+    "variants/cpu-int4/inference/generation_config.json": "4371a906cb4fb6ff3fdc71124a5e3566d034c7d074e7783be28e515e8a0dc954",
+    "variants/cpu-int4/inference/merger_lora_fpin_fpout.onnx": "953a8ad4704235c19ec06150788107921ba7f2578a5880a32cd59b473544712c",
+    "variants/cpu-int4/inference/merges.txt": "0b54e8aa4e53d5383e2e4bc635a56b43f9647f7b13832d5d9ecd8f82dac4f510",
+    "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin": "7b39156f3aa3b98af963314a0c929323b3f9c30a6f2a8c697f0d193347b50d1c",
+    "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin.sha256": "ec140a1d0028048b91e0fdd44f66565801f52335d366de65485f8add9554add7",
+    "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin": "8ed6b16e6734ec6f1ad820996737215dbb83716ceb563a58dd7bb35e15879ad3",
+    "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin.sha256": "9a6593e48f71aae9909953913aefe5cbc498ea86bab4ccdf554735b8215d8ee3",
+    "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin": "94bdab4e61ae33dc66b429b920fab629c3dd3bd956bcb88401c96a0e738707b5",
+    "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin.sha256": "ac7e4a72a1410276481c7bd69086bd32b0192d7ea18008382e5fd185201b42a8",
+    "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin": "c15669af396b6f627437a2339139ee2a53281069e45195cbcfe14663a487bbca",
+    "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin.sha256": "126b2ff6f4dc335a91e1b9b1b09b933e5cf6f49a382569632523a668eb71ee9a",
+    "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin": "989399d7ccb6da9feeac544e39a212f31e38b258c9609fcc010579761e3a30bc",
+    "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin.sha256": "9175a001d7cd3602e42eef69c7f8514b8a2f95b6fc8109d2c0d0576d5230e16b",
+    "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin": "b4f0b497896bf170e20e91deb370eb6179408089267dcae5c86553825c60d794",
+    "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin.sha256": "1f577f3e9adcab5ef3b0d1a3a1f6e2e518d7600cf44d600eb44b66d30a470bd0",
+    "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin": "a2ad50b46121b83c252f7b15e49f6a5ec47841344362c5e729396aace3cf887f",
+    "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin.sha256": "7ab3de6de7f781be5091cd6485b7da0ae1dab78f1145205a7083975c3557b70c",
+    "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin": "42ff8012f3acd8ce3ae86756ca1a03e782f65c87d3a901e5582037456a580681",
+    "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin.sha256": "e1ea5872eb7de99c02992c30ac948e50b31f9e0ac98bdf90fed10b898ae53212",
+    "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin": "1e8e94be6d0fbe3a06405195c8333949e6b515949fdfcde7f05c9a334bbe1848",
+    "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin.sha256": "c25eb763513b9d7f64b9e9be686451a4b6b4ee77d98b05add0a675afdd4c2dc8",
+    "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin": "25f2cd82eab7d258b1587caa5c9d59c3d239f7a3f08664f73cdfd94d334fd9cc",
+    "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin.sha256": "4bc58bf289758d8827f23ec3819000ba3a7d3b7654529571027b1e547fae70cf",
+    "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin": "f215d0a19f09b6ee6ae9629e2e3510076d50c5a79cdbfe807f58ac6a3ef20886",
+    "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin.sha256": "0c952131991d0607fa1898f031781261af0ba2d9d0de3f599944f52a87080579",
+    "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin": "fedf9da06f5adeb81d209f43c48ae69ac4eeac5214bdf90805f110493b4290c6",
+    "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin.sha256": "199e5d40de0f205581dcba68c4669f65288b6e401b301699e94e138a93cda6b5",
+    "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin": "d2a092ddaec7d726034c9aec2ccf5b5e49f7ae181eb249e73c9a525882ec1c3e",
+    "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin.sha256": "7fcc7c57ff369691616f888cde0f611f94a17534811e9e280a0c1d899edaa57c",
+    "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin": "055591adf01f2a965b1f92c1b0735429527956b8173243760ce31e0f66612c1c",
+    "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin.sha256": "67e413bbce04dad7130e7666e4e3a9fc173c41de598806db2d9c2a1c76e5bb30",
+    "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin": "d157255c424a9374de2d356c2f802e7cf52e73c3a80024c09512728ea42f47a6",
+    "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin.sha256": "c8270440c0ecf8f4a274efa4d581d424ce41f79c4db7460a2a5bdbc4c42aa8e5",
+    "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin": "ecf2158511b7e01a8137f922e1dd581c9a37d7d43192d40d19b09bc3e86c65e5",
+    "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin.sha256": "ca75931fbf41daa0a903632b7c8f77494a401dff9c4495de6f7c5bb30f76855b",
+    "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin": "762a55dcc7b5dd241b991ce83b68194a0ce88cfacd80fb16e4589189a47642b5",
+    "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin.sha256": "2db56c62e55dc1b4a1e06155bad608da35fbdcfbf49c85fb3e2817e8bfe73ea0",
+    "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin": "689033515724e0732230e084d38a1ba1c58a875d0ce569d9752c6c3139d56aeb",
+    "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin.sha256": "74b2882009096a52be4355a1621fb359ce5855482d9d0b1d6362efc25b6cd4be",
+    "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin": "86c162ec1d47015a2bcbd03bed790f74c912ffb6ac708c84c30007f308afa064",
+    "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin.sha256": "ede030b57df6f88927bed58b55b68e1a42eeeeadba676a183c8797f30aeb5955",
+    "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin": "a06945eae9c705524374286ff0c3f7872f0b7528804ee663c3b0b06f7c75d7c9",
+    "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin.sha256": "94714815fa72cd2a8738af9681091e84b58f43f82af2c075627bc4f92c4f0cb4",
+    "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin": "d9dac16f51339d395aa43d253e8caf19836865f8950ac4e9678c151f08ad5b30",
+    "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin.sha256": "582c21cf55cd0dee31b092020d31722564881a826ff35e1e5c5ec603a179ff10",
+    "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin": "cfa629ee97ac86104f939cc70ec7d5f527f5c345c63d5fe658e4c6495e755b52",
+    "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin.sha256": "5592716c5dee3f91da4cc54df057eb7b59ac46cd5574f323dc71efa6abd27fde",
+    "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin": "d427624d55a7de3ddb511493f4f67c58d78d51348e2b0a0487ba162b1e83aa3b",
+    "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin.sha256": "19cec7a719f6285a2b3526deb6cf11fce408be42f8598acf2ba953e97962ac5b",
+    "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin": "6bd9d3bfb4b15df17c520643aabcb3e4a94d085817099f8232cca7d71eddc0e3",
+    "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin.sha256": "acde77af27cfb96b809fc7f4fe5ebde99a9aabcd3ed1e2f5161daec1d3e2852f",
+    "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin": "98fe918e66abc0fd018287d7048afede2324d0d5669f900e38aa591f8bb17b6b",
+    "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin.sha256": "2aefa6e6bc0c5720d19eafafdf222eb97477f258995f37930925dcdcecb4bcbf",
+    "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin": "ac89ed9f33ad68ba97974985be51678275609c9841def0b6f50f688b613c035e",
+    "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin.sha256": "2a457fe40dec676ef46710bb6c289510180f867d4681c9baf42abdca9d398516",
+    "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin": "dcd4e16c50341f315fd6f564a39a3c86e03ffe6411d9963347e732d964ba6bd9",
+    "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin.sha256": "e64b0632f156215b741a818588514dfd4836996d7e4d71b69fda69fa75d54c3b",
+    "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin": "ecfba7aa07526af1bd9e3ee0347bc013413310ca55557e11c455ca53551dd573",
+    "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin.sha256": "837668db5f87d8a0c03da3be62942ce8ea9769dd90291146cb07db1dc60773f7",
+    "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin": "25683155e292658de393457949d58648dc2042755a0fb06c1eee5cf1d8bdb51d",
+    "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin.sha256": "b6a3bc1a10e4d344b3cf704ced560d985777c6fbd1ab0985be44fc05336c34e0",
+    "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin": "ca151a2092081042cd4719ecfbc6e394ac11ebb4d2b2c205bf3e9c23dcf2a4b8",
+    "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin.sha256": "387a61de7ebf0fd399fadefd3c99e3cb5e015114df9fcc5a0bd0856d3e79cc07",
+    "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin": "509e9ee37b74bfcacbb3f2464c05500ac604956e797f178bac63fb152efdbacf",
+    "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin.sha256": "00554704a7f438685504ac39788971cd63c900e892c3f7c7dbafc4a10ab48a2b",
+    "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin": "4ff27a2f91f1a03ab82ec06d7b99f5f0043ca396edd6c72614f39fd79050d173",
+    "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin.sha256": "c46aab01a82062baaca8648be10ba9a8fc2e2e054894fb50f3ab6c36b99d4c82",
+    "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin": "66d7e0eb02a410ea86fe6a0cfa3470692dad5635469622bcff0daefb62786fcd",
+    "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin.sha256": "2a29b357938273a816fbf5617231e3feed412cf5610cc7a80a36e170537af6cb",
+    "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin": "6a1eb7d077845a43e913e488d2d8a1f6f10ff44834a6f50a84400db0876d99da",
+    "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin.sha256": "b3ddeb2c83e3333085bc37df3a91d60e8d2670a63b85f47161d6f1716c965f7a",
+    "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin": "14d26ce08ef5a35f786023753948b3b7e8f38f89e6ed40ea5c5f221cefc38855",
+    "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin.sha256": "04db932bb7004b91da0b39bc85914dccb046e36a94056d2e1e8aceb366fda5bc",
+    "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin": "97376e53bd848b9ffc8a896e8656fe48a0c797b7279caee0bdf08971cde2882c",
+    "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin.sha256": "e9802cf6dc1a9c1808188f56dcf04fafaccfeb7043af50cebec273f829f2bfc5",
+    "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin": "0c8d49452806fb79719e90dc128b23a9e625151a7f160d551827e655c2ec4d46",
+    "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin.sha256": "de02163b223e71fcee66631dda7b1606d7b6fa15ed98a5bc32dcce4e7a518c93",
+    "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin": "97bfdc814ccaf451accb197d0d2d5c6de297155f331b13a4240d6d322b943a35",
+    "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin.sha256": "e1c6e7ecdc645e64a21aa19451b6f551011ca9c7c1374b4eaae484c6b4fe7bbd",
+    "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin": "3f84b8046f72e0c8955876cff0e9b5ed806d73caa9197e2bf57dbe02c112610f",
+    "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin.sha256": "c393705e520cf20b729a0ffda78ab214dfa9af1050863172bfa829ed66b2cc57",
+    "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin": "6da9fb6ec667392ea3a132714a370706880338e83fd4523dd451b9797f4e67a4",
+    "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin.sha256": "59fff04488a140fdee878062781def8034e5bf7fd23462bf5b4c0a75d42677a1",
+    "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin": "a010ede7984fbca031ab537cfc24455574a500b8379c2f9b8f81e57ef259976d",
+    "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin.sha256": "e5bab8d44a9ce0d502e2f3e1b709b5cbdad85d5865143c7720aa887257f5dac0",
+    "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin": "1dbd6d16da61ed25eb3c263282ac0775ca501830bda0efba4d9c13502e2e8cbe",
+    "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin.sha256": "e9e8c06d674aa15a540ecfa011335b02e27c3f52868ad66f95677fbc285585c9",
+    "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin": "b424ac2256be48f58afac149027b0fb579793ea43a653793acb5d7cd0abcd7f5",
+    "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin.sha256": "3644eeb869b2f244bc282e8458f4c2167eefb90092d61c6ae98060f435d6dba6",
+    "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin": "1c896b6a0ecf8ea3bbccc306f660378457255375998752a7f57474406f3e4f95",
+    "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin.sha256": "85534d0bc1ca866698e739146105dfefabbabed7b133c4b7e53e2f02d596c1a2",
+    "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin": "71cf7352257bfcd9fccbffb5fa350d98323e4794768ababf8afbd4e39e1c8285",
+    "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin.sha256": "5325e382d642eb4c70739d47ab7f413cc4793ad49e79e13fd9df0846a4f7a63b",
+    "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin": "04381a9835ddd434e5d531e07b4d84712dfabd4a7b2e04aba4f6f9e0692326bd",
+    "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin.sha256": "b8e88afff25780573ee89e50ee1708ebd06bf3cc8e650a5479b9cbed27c2d36c",
+    "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin": "73b4501f46ced84a91e050c5c812908baa1012b2c333295f2e989ab0bc6338c8",
+    "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin.sha256": "3afa1dc940e1af4bd11ecb7f14c38684d71641dcab505be8ca0bdef7249dd7e2",
+    "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin": "bb0d8cb31f79de8b5401d72f93434b026ab2007ac8e10f20045ef978992d8582",
+    "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin.sha256": "59fc535b420685ed1ac18b4e96c9e26ac64165398979c58ea50f3bad42d9a48a",
+    "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin": "52dd2e60b2c45ef4bbf44a39dcfd5be10da4be7b0a38e8ac6dfcf03b7b10c150",
+    "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin.sha256": "e5d9f006df6ddc3cc30db6adaf92ea63504b7ccf51841292c0babb796e95eba8",
+    "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin": "0cd176a8516d9f0e7aaaa44330f754e69993e9bdc9bc6c0b44d3ff2e6d53700b",
+    "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin.sha256": "91a90576bee9e062d0d5fabdfbebb7a45dc198fada27b6d90c7586a91a9746cc",
+    "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin": "2f5908db877b92336621fbfd16b9a6cbe6614727d945a452075afd7bb5698048",
+    "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin.sha256": "0ef27c5dddda8cc94160eddbd7897dca76d300dc4cf58dc9dd98362397bb54d6",
+    "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin": "ce21f7f781c1236f0a3afed6ca7600334d852508775b008ce92f9dabe6bcb75d",
+    "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin.sha256": "200f604cb77c055267052a1180ec7500aa0de8161b0684859c9b2c70e2171a01",
+    "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin": "ba2f1647e37b838817f380e953f6c1f07f22b809ba0228be6729f6b094211ada",
+    "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin.sha256": "e56624d071b1d3da9235f30928cb91f77f7500106f5621fc239168f4eb7e0d99",
+    "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin": "e6a19c3ae78203b50588867d15107866e1a21ef84e47554355754affc17184a3",
+    "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin.sha256": "9844f251663abbf4624922d61532ffbe693a3373c313fb63881adf7da211c9d1",
+    "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin": "c4cd40c94f2db60761799958454c004d6f913f1e0f9655ef02582bc1dccca36b",
+    "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin.sha256": "ef4debe3e646dcf8842d67ff7cc32945aaa70c7358da3ddbda144543433120de",
+    "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin": "bce7793797e491c449b7aaae2885bc03a3440ee8e78decbb98057177a17d83cc",
+    "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin.sha256": "55f08d5391fb10577009631d5f9cdfa6d738dda3d0caced324e0f67f6b18931e",
+    "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin": "ee832125df5322b43de1fcc4e4f18fdb9435063f74e9322543e6e7d463735093",
+    "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin.sha256": "2e6d243b8890b1fe1764e6ecac7c1e738778b95c646815628ea996c6537d3015",
+    "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin": "7dd553f1a8ce7e78718e3d05d97b57d9b9d2c9641e51bd2a1cc3b7a512e33a4c",
+    "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin.sha256": "1cb24179eeee75b4d40497b2f4a146e429c966081389b45daf7860208540aa8f",
+    "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin": "4642c9f601a97a847efc4bbbf1f9e61c23e7ea976e0ee5f12c4d980088d8e7d1",
+    "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin.sha256": "283b5f821c97e66048233f6b78725b24078700d5ac74cf33c876ad1059f57476",
+    "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin": "b0ee8d76e5aa150bd720796d37d421b3493b1fed370add5f7d2b216cd8af15f0",
+    "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin.sha256": "519aaca3915491339232a94094bbb661e1c8581088e2d82bb326f5e7e8c3a863",
+    "variants/cpu-int4/inference/model.onnx": "e81e3d0a0715a540f339d2b04ffe9ff1a1364fd9d91c9abc4b0ca8b9b49ad013",
+    "variants/cpu-int4/inference/optimum_config.json": "1f669998e4e0c9058f04ecef48a465556fc13886b2c87d055f52fb38cf2d77c9",
+    "variants/cpu-int4/inference/special_tokens_map.json": "2b7379f3ae813529281a5c602bc5a11c1d4e0a99107aaa597fe936c1e813ca52",
+    "variants/cpu-int4/inference/tokenizer.json": "7d27c493c729a66ecefc837280b05d948b1ed50d130eebdbf911b1b36cf38ed7",
+    "variants/cpu-int4/inference/tokenizer_config.json": "5dbff37e2d5899f5f69d86116f80733e11a790f703fd59855e2aef5717533ac5",
+    "variants/cpu-int4/inference/vocab.json": "82b84012e3add4d01d12ba14442026e49b8cbbaead1f79ecf3d919784f82dc79",
+    "variants/cpu-int4/inference/weight_handoff_map.json": "88831f61b1b56af1f201264a78ab9e0608b1fc9164f993c17e54d3c053e8bdfc",
+    "variants/cpu-int4/train/checkpoint": "ab316c91f27c8f2a36d82368c92ece5a63c7c6caa07f6d359f271dc6f1f05cca",
+    "variants/cpu-int4/train/eval_model.onnx": "77995f679f70746df7245fc05ed23a909b4726b251ffa6ead81a19c6a6647a82",
+    "variants/cpu-int4/train/optimizer_model.onnx": "8ed3b548bce52bf25c2154b1452314b3c48d079e28c52822dd348034f59dc054",
+    "variants/cpu-int4/train/trainable_parameters.json": "832cefe4d44e90acb1982c00de6b4516eec0bb0d4c22cda46d4d8c12b5c468e0",
+    "variants/cpu-int4/train/training_config.json": "caba5efc4b69d3b979dd220432aa48fa57ffaa4e98ef323d5346f0b275be7542",
+    "variants/cpu-int4/train/training_model.onnx": "ee26af1bec0e22928be4c7496518c1a38eebc7259e7369ed59e18365b1c8a602"
+  },
+  "supportedTasks": [
+    "text-generation-with-past"
+  ],
+  "trainableParameterCount": 460800,
+  "trainingParameterCount": 135436911,
+  "transformersVersion": "4.57.6",
+  "trustRemoteCode": false,
+  "variants": [
+    {
+      "abi": null,
+      "executionProvider": "cpu",
+      "features": [
+        "core",
+        "inference",
+        "train",
+        "rag"
+      ],
+      "id": "cpu-int4",
+      "minimumAndroidApi": 28,
+      "paths": {
+        "embedding": "variants/cpu-int4/embedding",
+        "inference": "variants/cpu-int4/inference",
+        "tokenizer": "shared/tokenizer",
+        "train": "variants/cpu-int4/train"
+      },
+      "quantization": "int4",
+      "recommendedDeviceMemoryMb": null,
+      "supportedEngines": [
+        "native"
+      ],
+      "weightHandoff": "variants/cpu-int4/inference/weight_handoff_map.json"
+    }
+  ],
+  "weightHandoff": "variants/cpu-int4/inference/weight_handoff_map.json"
+}

optimum/export_report.json

@@ -0,0 +1,35 @@
+{
+  "androidRuntime": {
+    "minimumAndroidApi": 28,
+    "recommendedDeviceMemoryMb": null,
+    "requiredAbis": []
+  },
+  "architectures": [
+    "llama"
+  ],
+  "license": {
+    "baseModelWeights": null,
+    "framework": null,
+    "noticeFile": null
+  },
+  "mobiletransformersVersion": "0.1.0",
+  "onnxRuntimeTrainingVersion": "1.23.0+cpu",
+  "optimumOnnxVersion": "0.1.0",
+  "peftMethods": [
+    "lora"
+  ],
+  "quantization": [
+    "int4"
+  ],
+  "selectedTask": "text-generation-with-past",
+  "sourceParameterCount": 134515008,
+  "supportedTasks": [
+    "text-generation-with-past"
+  ],
+  "trainInferenceLossDeltaNats": 0.1461,
+  "trainableParameterCount": 460800,
+  "trainingParameterCount": 135436911,
+  "trainingQuantizedParameterCount": 121208983,
+  "transformersVersion": "4.57.6",
+  "trustRemoteCode": false
+}

optimum/supported_tasks.json

@@ -0,0 +1,3 @@
+[
+  "text-generation-with-past"
+]

shared/tokenizer/chat_template.jinja

@@ -0,0 +1,6 @@
+{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
+You are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>
+' }}{% endif %}{{'<|im_start|>' + message['role'] + '
+' + message['content'] + '<|im_end|>' + '
+'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
+' }}{% endif %}

shared/tokenizer/mobiletransformers_tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+    "model": {
+        "bos_token_id": 1,
+        "context_length": 8192,
+        "num_attention_heads": 9,
+        "num_hidden_layers": 30,
+        "num_key_value_heads": 3,
+        "eos_token_id": 2,
+        "pad_token_id": 2,
+        "type": "llama",
+        "vocab_size": 49152
+    }
+}

shared/tokenizer/special_tokens_map.json

@@ -0,0 +1,34 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "bos_token": {
+    "content": "<|im_start|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

shared/tokenizer/tokenizer_config.json

@@ -0,0 +1,154 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<repo_name>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "<reponame>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "<file_sep>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "<filename>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "7": {
+      "content": "<gh_stars>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "8": {
+      "content": "<issue_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "9": {
+      "content": "<issue_comment>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "10": {
+      "content": "<issue_closed>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "11": {
+      "content": "<jupyter_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "12": {
+      "content": "<jupyter_text>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "13": {
+      "content": "<jupyter_code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "14": {
+      "content": "<jupyter_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "15": {
+      "content": "<jupyter_script>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "16": {
+      "content": "<empty_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "bos_token": "<|im_start|>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "extra_special_tokens": {},
+  "model_max_length": 8192,
+  "pad_token": "<|im_end|>",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>",
+  "vocab_size": 49152
+}

variants/cpu-int4/checksums.json

@@ -0,0 +1,148 @@
+{
+  "variants/cpu-int4/checksums.json": "6c205ecff7f07d8c3117fe0ce7ecef194741ef9e2fb34e22c95150fc9e7f1367",
+  "variants/cpu-int4/embedding/embedding_model.onnx": "7c46645e1dcec3d3a27d80f21fcad81ab307317bf3d0580eb8ea6779fa83cc96",
+  "variants/cpu-int4/embedding/rag_config.json": "f90d993929e9e7b44c0cbe4518c72e7c64979a527a9d91a766dc967dea95de16",
+  "variants/cpu-int4/embedding/tokenizer/special_tokens_map.json": "5d5b662e421ea9fac075174bb0688ee0d9431699900b90662acd44b2a350503a",
+  "variants/cpu-int4/embedding/tokenizer/tokenizer.json": "da0e79933b9ed51798a3ae27893d3c5fa4a201126cef75586296df9b4d2c62a0",
+  "variants/cpu-int4/embedding/tokenizer/tokenizer_config.json": "ccb4eb21a03e1442ee5c3f85431b9c307960a04579942537d74778dc8080a48c",
+  "variants/cpu-int4/inference/chat_template.jinja": "872be49dbb638044ad01b60388f48d469ff2980e5f0dccdc22ec907db54d0788",
+  "variants/cpu-int4/inference/config.json": "2f679c28417584f0b3a2d79542dd7da73f27cf927889434afa6b1731a091ef8e",
+  "variants/cpu-int4/inference/frozen_base.onnx.data": "12f4d16fce65a2186843b4b435bced33506c16b0f9a43dc1fad5388d99576548",
+  "variants/cpu-int4/inference/genai_config.json": "4a80b27a1e0bcde8917b7b15c2234813f48e1b894b7768f7fa68e07ff7e558d9",
+  "variants/cpu-int4/inference/generation_config.json": "4371a906cb4fb6ff3fdc71124a5e3566d034c7d074e7783be28e515e8a0dc954",
+  "variants/cpu-int4/inference/merger_lora_fpin_fpout.onnx": "953a8ad4704235c19ec06150788107921ba7f2578a5880a32cd59b473544712c",
+  "variants/cpu-int4/inference/merges.txt": "0b54e8aa4e53d5383e2e4bc635a56b43f9647f7b13832d5d9ecd8f82dac4f510",
+  "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin": "7b39156f3aa3b98af963314a0c929323b3f9c30a6f2a8c697f0d193347b50d1c",
+  "variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin.sha256": "ec140a1d0028048b91e0fdd44f66565801f52335d366de65485f8add9554add7",
+  "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin": "8ed6b16e6734ec6f1ad820996737215dbb83716ceb563a58dd7bb35e15879ad3",
+  "variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin.sha256": "9a6593e48f71aae9909953913aefe5cbc498ea86bab4ccdf554735b8215d8ee3",
+  "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin": "94bdab4e61ae33dc66b429b920fab629c3dd3bd956bcb88401c96a0e738707b5",
+  "variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin.sha256": "ac7e4a72a1410276481c7bd69086bd32b0192d7ea18008382e5fd185201b42a8",
+  "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin": "c15669af396b6f627437a2339139ee2a53281069e45195cbcfe14663a487bbca",
+  "variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin.sha256": "126b2ff6f4dc335a91e1b9b1b09b933e5cf6f49a382569632523a668eb71ee9a",
+  "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin": "989399d7ccb6da9feeac544e39a212f31e38b258c9609fcc010579761e3a30bc",
+  "variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin.sha256": "9175a001d7cd3602e42eef69c7f8514b8a2f95b6fc8109d2c0d0576d5230e16b",
+  "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin": "b4f0b497896bf170e20e91deb370eb6179408089267dcae5c86553825c60d794",
+  "variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin.sha256": "1f577f3e9adcab5ef3b0d1a3a1f6e2e518d7600cf44d600eb44b66d30a470bd0",
+  "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin": "a2ad50b46121b83c252f7b15e49f6a5ec47841344362c5e729396aace3cf887f",
+  "variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin.sha256": "7ab3de6de7f781be5091cd6485b7da0ae1dab78f1145205a7083975c3557b70c",
+  "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin": "42ff8012f3acd8ce3ae86756ca1a03e782f65c87d3a901e5582037456a580681",
+  "variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin.sha256": "e1ea5872eb7de99c02992c30ac948e50b31f9e0ac98bdf90fed10b898ae53212",
+  "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin": "1e8e94be6d0fbe3a06405195c8333949e6b515949fdfcde7f05c9a334bbe1848",
+  "variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin.sha256": "c25eb763513b9d7f64b9e9be686451a4b6b4ee77d98b05add0a675afdd4c2dc8",
+  "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin": "25f2cd82eab7d258b1587caa5c9d59c3d239f7a3f08664f73cdfd94d334fd9cc",
+  "variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin.sha256": "4bc58bf289758d8827f23ec3819000ba3a7d3b7654529571027b1e547fae70cf",
+  "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin": "f215d0a19f09b6ee6ae9629e2e3510076d50c5a79cdbfe807f58ac6a3ef20886",
+  "variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin.sha256": "0c952131991d0607fa1898f031781261af0ba2d9d0de3f599944f52a87080579",
+  "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin": "fedf9da06f5adeb81d209f43c48ae69ac4eeac5214bdf90805f110493b4290c6",
+  "variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin.sha256": "199e5d40de0f205581dcba68c4669f65288b6e401b301699e94e138a93cda6b5",
+  "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin": "d2a092ddaec7d726034c9aec2ccf5b5e49f7ae181eb249e73c9a525882ec1c3e",
+  "variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin.sha256": "7fcc7c57ff369691616f888cde0f611f94a17534811e9e280a0c1d899edaa57c",
+  "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin": "055591adf01f2a965b1f92c1b0735429527956b8173243760ce31e0f66612c1c",
+  "variants/cpu-int4/inference/model.layers.14.self_attn.v_proj.MatMul.weight.bin.sha256": "67e413bbce04dad7130e7666e4e3a9fc173c41de598806db2d9c2a1c76e5bb30",
+  "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin": "d157255c424a9374de2d356c2f802e7cf52e73c3a80024c09512728ea42f47a6",
+  "variants/cpu-int4/inference/model.layers.15.self_attn.q_proj.MatMul.weight.bin.sha256": "c8270440c0ecf8f4a274efa4d581d424ce41f79c4db7460a2a5bdbc4c42aa8e5",
+  "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin": "ecf2158511b7e01a8137f922e1dd581c9a37d7d43192d40d19b09bc3e86c65e5",
+  "variants/cpu-int4/inference/model.layers.15.self_attn.v_proj.MatMul.weight.bin.sha256": "ca75931fbf41daa0a903632b7c8f77494a401dff9c4495de6f7c5bb30f76855b",
+  "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin": "762a55dcc7b5dd241b991ce83b68194a0ce88cfacd80fb16e4589189a47642b5",
+  "variants/cpu-int4/inference/model.layers.16.self_attn.q_proj.MatMul.weight.bin.sha256": "2db56c62e55dc1b4a1e06155bad608da35fbdcfbf49c85fb3e2817e8bfe73ea0",
+  "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin": "689033515724e0732230e084d38a1ba1c58a875d0ce569d9752c6c3139d56aeb",
+  "variants/cpu-int4/inference/model.layers.16.self_attn.v_proj.MatMul.weight.bin.sha256": "74b2882009096a52be4355a1621fb359ce5855482d9d0b1d6362efc25b6cd4be",
+  "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin": "86c162ec1d47015a2bcbd03bed790f74c912ffb6ac708c84c30007f308afa064",
+  "variants/cpu-int4/inference/model.layers.17.self_attn.q_proj.MatMul.weight.bin.sha256": "ede030b57df6f88927bed58b55b68e1a42eeeeadba676a183c8797f30aeb5955",
+  "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin": "a06945eae9c705524374286ff0c3f7872f0b7528804ee663c3b0b06f7c75d7c9",
+  "variants/cpu-int4/inference/model.layers.17.self_attn.v_proj.MatMul.weight.bin.sha256": "94714815fa72cd2a8738af9681091e84b58f43f82af2c075627bc4f92c4f0cb4",
+  "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin": "d9dac16f51339d395aa43d253e8caf19836865f8950ac4e9678c151f08ad5b30",
+  "variants/cpu-int4/inference/model.layers.18.self_attn.q_proj.MatMul.weight.bin.sha256": "582c21cf55cd0dee31b092020d31722564881a826ff35e1e5c5ec603a179ff10",
+  "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin": "cfa629ee97ac86104f939cc70ec7d5f527f5c345c63d5fe658e4c6495e755b52",
+  "variants/cpu-int4/inference/model.layers.18.self_attn.v_proj.MatMul.weight.bin.sha256": "5592716c5dee3f91da4cc54df057eb7b59ac46cd5574f323dc71efa6abd27fde",
+  "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin": "d427624d55a7de3ddb511493f4f67c58d78d51348e2b0a0487ba162b1e83aa3b",
+  "variants/cpu-int4/inference/model.layers.19.self_attn.q_proj.MatMul.weight.bin.sha256": "19cec7a719f6285a2b3526deb6cf11fce408be42f8598acf2ba953e97962ac5b",
+  "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin": "6bd9d3bfb4b15df17c520643aabcb3e4a94d085817099f8232cca7d71eddc0e3",
+  "variants/cpu-int4/inference/model.layers.19.self_attn.v_proj.MatMul.weight.bin.sha256": "acde77af27cfb96b809fc7f4fe5ebde99a9aabcd3ed1e2f5161daec1d3e2852f",
+  "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin": "98fe918e66abc0fd018287d7048afede2324d0d5669f900e38aa591f8bb17b6b",
+  "variants/cpu-int4/inference/model.layers.2.self_attn.q_proj.MatMul.weight.bin.sha256": "2aefa6e6bc0c5720d19eafafdf222eb97477f258995f37930925dcdcecb4bcbf",
+  "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin": "ac89ed9f33ad68ba97974985be51678275609c9841def0b6f50f688b613c035e",
+  "variants/cpu-int4/inference/model.layers.2.self_attn.v_proj.MatMul.weight.bin.sha256": "2a457fe40dec676ef46710bb6c289510180f867d4681c9baf42abdca9d398516",
+  "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin": "dcd4e16c50341f315fd6f564a39a3c86e03ffe6411d9963347e732d964ba6bd9",
+  "variants/cpu-int4/inference/model.layers.20.self_attn.q_proj.MatMul.weight.bin.sha256": "e64b0632f156215b741a818588514dfd4836996d7e4d71b69fda69fa75d54c3b",
+  "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin": "ecfba7aa07526af1bd9e3ee0347bc013413310ca55557e11c455ca53551dd573",
+  "variants/cpu-int4/inference/model.layers.20.self_attn.v_proj.MatMul.weight.bin.sha256": "837668db5f87d8a0c03da3be62942ce8ea9769dd90291146cb07db1dc60773f7",
+  "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin": "25683155e292658de393457949d58648dc2042755a0fb06c1eee5cf1d8bdb51d",
+  "variants/cpu-int4/inference/model.layers.21.self_attn.q_proj.MatMul.weight.bin.sha256": "b6a3bc1a10e4d344b3cf704ced560d985777c6fbd1ab0985be44fc05336c34e0",
+  "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin": "ca151a2092081042cd4719ecfbc6e394ac11ebb4d2b2c205bf3e9c23dcf2a4b8",
+  "variants/cpu-int4/inference/model.layers.21.self_attn.v_proj.MatMul.weight.bin.sha256": "387a61de7ebf0fd399fadefd3c99e3cb5e015114df9fcc5a0bd0856d3e79cc07",
+  "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin": "509e9ee37b74bfcacbb3f2464c05500ac604956e797f178bac63fb152efdbacf",
+  "variants/cpu-int4/inference/model.layers.22.self_attn.q_proj.MatMul.weight.bin.sha256": "00554704a7f438685504ac39788971cd63c900e892c3f7c7dbafc4a10ab48a2b",
+  "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin": "4ff27a2f91f1a03ab82ec06d7b99f5f0043ca396edd6c72614f39fd79050d173",
+  "variants/cpu-int4/inference/model.layers.22.self_attn.v_proj.MatMul.weight.bin.sha256": "c46aab01a82062baaca8648be10ba9a8fc2e2e054894fb50f3ab6c36b99d4c82",
+  "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin": "66d7e0eb02a410ea86fe6a0cfa3470692dad5635469622bcff0daefb62786fcd",
+  "variants/cpu-int4/inference/model.layers.23.self_attn.q_proj.MatMul.weight.bin.sha256": "2a29b357938273a816fbf5617231e3feed412cf5610cc7a80a36e170537af6cb",
+  "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin": "6a1eb7d077845a43e913e488d2d8a1f6f10ff44834a6f50a84400db0876d99da",
+  "variants/cpu-int4/inference/model.layers.23.self_attn.v_proj.MatMul.weight.bin.sha256": "b3ddeb2c83e3333085bc37df3a91d60e8d2670a63b85f47161d6f1716c965f7a",
+  "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin": "14d26ce08ef5a35f786023753948b3b7e8f38f89e6ed40ea5c5f221cefc38855",
+  "variants/cpu-int4/inference/model.layers.24.self_attn.q_proj.MatMul.weight.bin.sha256": "04db932bb7004b91da0b39bc85914dccb046e36a94056d2e1e8aceb366fda5bc",
+  "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin": "97376e53bd848b9ffc8a896e8656fe48a0c797b7279caee0bdf08971cde2882c",
+  "variants/cpu-int4/inference/model.layers.24.self_attn.v_proj.MatMul.weight.bin.sha256": "e9802cf6dc1a9c1808188f56dcf04fafaccfeb7043af50cebec273f829f2bfc5",
+  "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin": "0c8d49452806fb79719e90dc128b23a9e625151a7f160d551827e655c2ec4d46",
+  "variants/cpu-int4/inference/model.layers.25.self_attn.q_proj.MatMul.weight.bin.sha256": "de02163b223e71fcee66631dda7b1606d7b6fa15ed98a5bc32dcce4e7a518c93",
+  "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin": "97bfdc814ccaf451accb197d0d2d5c6de297155f331b13a4240d6d322b943a35",
+  "variants/cpu-int4/inference/model.layers.25.self_attn.v_proj.MatMul.weight.bin.sha256": "e1c6e7ecdc645e64a21aa19451b6f551011ca9c7c1374b4eaae484c6b4fe7bbd",
+  "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin": "3f84b8046f72e0c8955876cff0e9b5ed806d73caa9197e2bf57dbe02c112610f",
+  "variants/cpu-int4/inference/model.layers.26.self_attn.q_proj.MatMul.weight.bin.sha256": "c393705e520cf20b729a0ffda78ab214dfa9af1050863172bfa829ed66b2cc57",
+  "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin": "6da9fb6ec667392ea3a132714a370706880338e83fd4523dd451b9797f4e67a4",
+  "variants/cpu-int4/inference/model.layers.26.self_attn.v_proj.MatMul.weight.bin.sha256": "59fff04488a140fdee878062781def8034e5bf7fd23462bf5b4c0a75d42677a1",
+  "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin": "a010ede7984fbca031ab537cfc24455574a500b8379c2f9b8f81e57ef259976d",
+  "variants/cpu-int4/inference/model.layers.27.self_attn.q_proj.MatMul.weight.bin.sha256": "e5bab8d44a9ce0d502e2f3e1b709b5cbdad85d5865143c7720aa887257f5dac0",
+  "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin": "1dbd6d16da61ed25eb3c263282ac0775ca501830bda0efba4d9c13502e2e8cbe",
+  "variants/cpu-int4/inference/model.layers.27.self_attn.v_proj.MatMul.weight.bin.sha256": "e9e8c06d674aa15a540ecfa011335b02e27c3f52868ad66f95677fbc285585c9",
+  "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin": "b424ac2256be48f58afac149027b0fb579793ea43a653793acb5d7cd0abcd7f5",
+  "variants/cpu-int4/inference/model.layers.28.self_attn.q_proj.MatMul.weight.bin.sha256": "3644eeb869b2f244bc282e8458f4c2167eefb90092d61c6ae98060f435d6dba6",
+  "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin": "1c896b6a0ecf8ea3bbccc306f660378457255375998752a7f57474406f3e4f95",
+  "variants/cpu-int4/inference/model.layers.28.self_attn.v_proj.MatMul.weight.bin.sha256": "85534d0bc1ca866698e739146105dfefabbabed7b133c4b7e53e2f02d596c1a2",
+  "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin": "71cf7352257bfcd9fccbffb5fa350d98323e4794768ababf8afbd4e39e1c8285",
+  "variants/cpu-int4/inference/model.layers.29.self_attn.q_proj.MatMul.weight.bin.sha256": "5325e382d642eb4c70739d47ab7f413cc4793ad49e79e13fd9df0846a4f7a63b",
+  "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin": "04381a9835ddd434e5d531e07b4d84712dfabd4a7b2e04aba4f6f9e0692326bd",
+  "variants/cpu-int4/inference/model.layers.29.self_attn.v_proj.MatMul.weight.bin.sha256": "b8e88afff25780573ee89e50ee1708ebd06bf3cc8e650a5479b9cbed27c2d36c",
+  "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin": "73b4501f46ced84a91e050c5c812908baa1012b2c333295f2e989ab0bc6338c8",
+  "variants/cpu-int4/inference/model.layers.3.self_attn.q_proj.MatMul.weight.bin.sha256": "3afa1dc940e1af4bd11ecb7f14c38684d71641dcab505be8ca0bdef7249dd7e2",
+  "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin": "bb0d8cb31f79de8b5401d72f93434b026ab2007ac8e10f20045ef978992d8582",
+  "variants/cpu-int4/inference/model.layers.3.self_attn.v_proj.MatMul.weight.bin.sha256": "59fc535b420685ed1ac18b4e96c9e26ac64165398979c58ea50f3bad42d9a48a",
+  "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin": "52dd2e60b2c45ef4bbf44a39dcfd5be10da4be7b0a38e8ac6dfcf03b7b10c150",
+  "variants/cpu-int4/inference/model.layers.4.self_attn.q_proj.MatMul.weight.bin.sha256": "e5d9f006df6ddc3cc30db6adaf92ea63504b7ccf51841292c0babb796e95eba8",
+  "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin": "0cd176a8516d9f0e7aaaa44330f754e69993e9bdc9bc6c0b44d3ff2e6d53700b",
+  "variants/cpu-int4/inference/model.layers.4.self_attn.v_proj.MatMul.weight.bin.sha256": "91a90576bee9e062d0d5fabdfbebb7a45dc198fada27b6d90c7586a91a9746cc",
+  "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin": "2f5908db877b92336621fbfd16b9a6cbe6614727d945a452075afd7bb5698048",
+  "variants/cpu-int4/inference/model.layers.5.self_attn.q_proj.MatMul.weight.bin.sha256": "0ef27c5dddda8cc94160eddbd7897dca76d300dc4cf58dc9dd98362397bb54d6",
+  "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin": "ce21f7f781c1236f0a3afed6ca7600334d852508775b008ce92f9dabe6bcb75d",
+  "variants/cpu-int4/inference/model.layers.5.self_attn.v_proj.MatMul.weight.bin.sha256": "200f604cb77c055267052a1180ec7500aa0de8161b0684859c9b2c70e2171a01",
+  "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin": "ba2f1647e37b838817f380e953f6c1f07f22b809ba0228be6729f6b094211ada",
+  "variants/cpu-int4/inference/model.layers.6.self_attn.q_proj.MatMul.weight.bin.sha256": "e56624d071b1d3da9235f30928cb91f77f7500106f5621fc239168f4eb7e0d99",
+  "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin": "e6a19c3ae78203b50588867d15107866e1a21ef84e47554355754affc17184a3",
+  "variants/cpu-int4/inference/model.layers.6.self_attn.v_proj.MatMul.weight.bin.sha256": "9844f251663abbf4624922d61532ffbe693a3373c313fb63881adf7da211c9d1",
+  "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin": "c4cd40c94f2db60761799958454c004d6f913f1e0f9655ef02582bc1dccca36b",
+  "variants/cpu-int4/inference/model.layers.7.self_attn.q_proj.MatMul.weight.bin.sha256": "ef4debe3e646dcf8842d67ff7cc32945aaa70c7358da3ddbda144543433120de",
+  "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin": "bce7793797e491c449b7aaae2885bc03a3440ee8e78decbb98057177a17d83cc",
+  "variants/cpu-int4/inference/model.layers.7.self_attn.v_proj.MatMul.weight.bin.sha256": "55f08d5391fb10577009631d5f9cdfa6d738dda3d0caced324e0f67f6b18931e",
+  "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin": "ee832125df5322b43de1fcc4e4f18fdb9435063f74e9322543e6e7d463735093",
+  "variants/cpu-int4/inference/model.layers.8.self_attn.q_proj.MatMul.weight.bin.sha256": "2e6d243b8890b1fe1764e6ecac7c1e738778b95c646815628ea996c6537d3015",
+  "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin": "7dd553f1a8ce7e78718e3d05d97b57d9b9d2c9641e51bd2a1cc3b7a512e33a4c",
+  "variants/cpu-int4/inference/model.layers.8.self_attn.v_proj.MatMul.weight.bin.sha256": "1cb24179eeee75b4d40497b2f4a146e429c966081389b45daf7860208540aa8f",
+  "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin": "4642c9f601a97a847efc4bbbf1f9e61c23e7ea976e0ee5f12c4d980088d8e7d1",
+  "variants/cpu-int4/inference/model.layers.9.self_attn.q_proj.MatMul.weight.bin.sha256": "283b5f821c97e66048233f6b78725b24078700d5ac74cf33c876ad1059f57476",
+  "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin": "b0ee8d76e5aa150bd720796d37d421b3493b1fed370add5f7d2b216cd8af15f0",
+  "variants/cpu-int4/inference/model.layers.9.self_attn.v_proj.MatMul.weight.bin.sha256": "519aaca3915491339232a94094bbb661e1c8581088e2d82bb326f5e7e8c3a863",
+  "variants/cpu-int4/inference/model.onnx": "e81e3d0a0715a540f339d2b04ffe9ff1a1364fd9d91c9abc4b0ca8b9b49ad013",
+  "variants/cpu-int4/inference/optimum_config.json": "1f669998e4e0c9058f04ecef48a465556fc13886b2c87d055f52fb38cf2d77c9",
+  "variants/cpu-int4/inference/special_tokens_map.json": "2b7379f3ae813529281a5c602bc5a11c1d4e0a99107aaa597fe936c1e813ca52",
+  "variants/cpu-int4/inference/tokenizer.json": "7d27c493c729a66ecefc837280b05d948b1ed50d130eebdbf911b1b36cf38ed7",
+  "variants/cpu-int4/inference/tokenizer_config.json": "5dbff37e2d5899f5f69d86116f80733e11a790f703fd59855e2aef5717533ac5",
+  "variants/cpu-int4/inference/vocab.json": "82b84012e3add4d01d12ba14442026e49b8cbbaead1f79ecf3d919784f82dc79",
+  "variants/cpu-int4/inference/weight_handoff_map.json": "88831f61b1b56af1f201264a78ab9e0608b1fc9164f993c17e54d3c053e8bdfc",
+  "variants/cpu-int4/train/checkpoint": "ab316c91f27c8f2a36d82368c92ece5a63c7c6caa07f6d359f271dc6f1f05cca",
+  "variants/cpu-int4/train/eval_model.onnx": "77995f679f70746df7245fc05ed23a909b4726b251ffa6ead81a19c6a6647a82",
+  "variants/cpu-int4/train/optimizer_model.onnx": "8ed3b548bce52bf25c2154b1452314b3c48d079e28c52822dd348034f59dc054",
+  "variants/cpu-int4/train/trainable_parameters.json": "832cefe4d44e90acb1982c00de6b4516eec0bb0d4c22cda46d4d8c12b5c468e0",
+  "variants/cpu-int4/train/training_config.json": "caba5efc4b69d3b979dd220432aa48fa57ffaa4e98ef323d5346f0b275be7542",
+  "variants/cpu-int4/train/training_model.onnx": "ee26af1bec0e22928be4c7496518c1a38eebc7259e7369ed59e18365b1c8a602"
+}

variants/cpu-int4/embedding/embedding_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c46645e1dcec3d3a27d80f21fcad81ab307317bf3d0580eb8ea6779fa83cc96
+size 90392834

variants/cpu-int4/embedding/rag_config.json

@@ -0,0 +1,13 @@
+{
+  "chunkOverlap": 50,
+  "chunkSize": 512,
+  "embeddingDimension": 384,
+  "embeddingModelId": "sentence-transformers/all-MiniLM-L6-v2",
+  "indexingMode": "precompute",
+  "maxTextLength": 1024,
+  "minScore": 0.0,
+  "onnxName": "embedding_model",
+  "repoName": "HuggingFaceTB__SmolLM2-135M-Instruct",
+  "searchType": "semantic",
+  "topK": 10
+}

variants/cpu-int4/embedding/tokenizer/special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

variants/cpu-int4/embedding/tokenizer/tokenizer_config.json

@@ -0,0 +1,65 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "max_length": 128,
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_to_multiple_of": null,
+  "pad_token": "[PAD]",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}

variants/cpu-int4/inference/chat_template.jinja

@@ -0,0 +1,6 @@
+{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
+You are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>
+' }}{% endif %}{{'<|im_start|>' + message['role'] + '
+' + message['content'] + '<|im_end|>' + '
+'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
+' }}{% endif %}

variants/cpu-int4/inference/config.json

@@ -0,0 +1,38 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 576,
+  "initializer_range": 0.041666666666666664,
+  "intermediate_size": 1536,
+  "is_llama_config": true,
+  "max_position_embeddings": 8192,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 9,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 3,
+  "pad_token_id": 2,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_interleaved": false,
+  "rope_scaling": null,
+  "rope_theta": 100000,
+  "tie_word_embeddings": true,
+  "transformers.js_config": {
+    "kv_cache_dtype": {
+      "fp16": "float16",
+      "q4f16": "float16"
+    }
+  },
+  "transformers_version": "4.57.6",
+  "use_cache": true,
+  "vocab_size": 49152
+}

variants/cpu-int4/inference/frozen_base.onnx.data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12f4d16fce65a2186843b4b435bced33506c16b0f9a43dc1fad5388d99576548
+size 598222080

variants/cpu-int4/inference/genai_config.json

@@ -0,0 +1,42 @@
+{
+  "model": {
+    "bos_token_id": 1,
+    "eos_token_id": 2,
+    "pad_token_id": 2,
+    "context_length": 8192,
+    "type": "llama",
+    "vocab_size": 49152,
+    "decoder": {
+      "session_options": {
+        "provider_options": []
+      },
+      "filename": "model.onnx",
+      "head_size": 64,
+      "hidden_size": 576,
+      "num_attention_heads": 9,
+      "num_key_value_heads": 3,
+      "num_hidden_layers": 30,
+      "inputs": {
+        "input_ids": "input_ids",
+        "attention_mask": "attention_mask",
+        "position_ids": "position_ids",
+        "past_key_names": "past_key_values.%d.key",
+        "past_value_names": "past_key_values.%d.value"
+      },
+      "outputs": {
+        "logits": "logits",
+        "present_key_names": "present.%d.key",
+        "present_value_names": "present.%d.value"
+      }
+    }
+  },
+  "search": {
+    "max_length": 8192,
+    "min_length": 0,
+    "do_sample": false,
+    "top_k": 1,
+    "top_p": 1.0,
+    "temperature": 1.0,
+    "repetition_penalty": 1.0
+  }
+}

variants/cpu-int4/inference/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 2,
+  "transformers_version": "4.57.6"
+}

variants/cpu-int4/inference/merger_lora_fpin_fpout.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:953a8ad4704235c19ec06150788107921ba7f2578a5880a32cd59b473544712c
+size 811

variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b39156f3aa3b98af963314a0c929323b3f9c30a6f2a8c697f0d193347b50d1c
+size 1327104

variants/cpu-int4/inference/model.layers.0.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+7b39156f3aa3b98af963314a0c929323b3f9c30a6f2a8c697f0d193347b50d1c

variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ed6b16e6734ec6f1ad820996737215dbb83716ceb563a58dd7bb35e15879ad3
+size 442368

variants/cpu-int4/inference/model.layers.0.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+8ed6b16e6734ec6f1ad820996737215dbb83716ceb563a58dd7bb35e15879ad3

variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94bdab4e61ae33dc66b429b920fab629c3dd3bd956bcb88401c96a0e738707b5
+size 1327104

variants/cpu-int4/inference/model.layers.1.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+94bdab4e61ae33dc66b429b920fab629c3dd3bd956bcb88401c96a0e738707b5

variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c15669af396b6f627437a2339139ee2a53281069e45195cbcfe14663a487bbca
+size 442368

variants/cpu-int4/inference/model.layers.1.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+c15669af396b6f627437a2339139ee2a53281069e45195cbcfe14663a487bbca

variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:989399d7ccb6da9feeac544e39a212f31e38b258c9609fcc010579761e3a30bc
+size 1327104

variants/cpu-int4/inference/model.layers.10.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+989399d7ccb6da9feeac544e39a212f31e38b258c9609fcc010579761e3a30bc

variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4f0b497896bf170e20e91deb370eb6179408089267dcae5c86553825c60d794
+size 442368

variants/cpu-int4/inference/model.layers.10.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+b4f0b497896bf170e20e91deb370eb6179408089267dcae5c86553825c60d794

variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2ad50b46121b83c252f7b15e49f6a5ec47841344362c5e729396aace3cf887f
+size 1327104

variants/cpu-int4/inference/model.layers.11.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+a2ad50b46121b83c252f7b15e49f6a5ec47841344362c5e729396aace3cf887f

variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42ff8012f3acd8ce3ae86756ca1a03e782f65c87d3a901e5582037456a580681
+size 442368

variants/cpu-int4/inference/model.layers.11.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+42ff8012f3acd8ce3ae86756ca1a03e782f65c87d3a901e5582037456a580681

variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e8e94be6d0fbe3a06405195c8333949e6b515949fdfcde7f05c9a334bbe1848
+size 1327104

variants/cpu-int4/inference/model.layers.12.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+1e8e94be6d0fbe3a06405195c8333949e6b515949fdfcde7f05c9a334bbe1848

variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25f2cd82eab7d258b1587caa5c9d59c3d239f7a3f08664f73cdfd94d334fd9cc
+size 442368

variants/cpu-int4/inference/model.layers.12.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+25f2cd82eab7d258b1587caa5c9d59c3d239f7a3f08664f73cdfd94d334fd9cc

variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f215d0a19f09b6ee6ae9629e2e3510076d50c5a79cdbfe807f58ac6a3ef20886
+size 1327104

variants/cpu-int4/inference/model.layers.13.self_attn.q_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+f215d0a19f09b6ee6ae9629e2e3510076d50c5a79cdbfe807f58ac6a3ef20886

variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fedf9da06f5adeb81d209f43c48ae69ac4eeac5214bdf90805f110493b4290c6
+size 442368

variants/cpu-int4/inference/model.layers.13.self_attn.v_proj.MatMul.weight.bin.sha256

@@ -0,0 +1 @@
+fedf9da06f5adeb81d209f43c48ae69ac4eeac5214bdf90805f110493b4290c6

variants/cpu-int4/inference/model.layers.14.self_attn.q_proj.MatMul.weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2a092ddaec7d726034c9aec2ccf5b5e49f7ae181eb249e73c9a525882ec1c3e
+size 1327104