Upload folder using huggingface_hub
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +3 -0
- GGUF_Q4_K_M_ADDED_FOUR_REPORT.json +31 -0
- GGUF_Q4_K_M_REPORT.json +49 -0
- README.md +315 -0
- REPORT.md +40 -0
- deepseek-coder/case.json +58 -0
- deepseek-coder/config-diff.json +47 -0
- deepseek-coder/gguf/metadata.json +96 -0
- deepseek-coder/gguf/model-Q4_K_M.gguf +0 -0
- deepseek-coder/gguf/reference.json +21 -0
- deepseek-coder/inputs.safetensors +3 -0
- deepseek-coder/package/config.json +30 -0
- deepseek-coder/package/generation_config.json +6 -0
- deepseek-coder/package/model.safetensors +3 -0
- deepseek-coder/provenance.json +19 -0
- deepseek-coder/reference/outputs.safetensors +3 -0
- deepseek-coder/shrunk-config.json +30 -0
- deepseek-coder/source-config.json +30 -0
- deepseek-coder/tokenizer/metadata.json +24 -0
- deepseek-coder/tokenizer/special_tokens_map.json +5 -0
- deepseek-coder/tokenizer/tokenizer.model +3 -0
- deepseek-coder/tokenizer/tokenizer_config.json +12 -0
- deepseek-coder/tokenizer/vocabulary.json +130 -0
- deepseek-coder/validation.json +72 -0
- hermes3-llama31/case.json +58 -0
- hermes3-llama31/config-diff.json +47 -0
- hermes3-llama31/gguf/metadata.json +96 -0
- hermes3-llama31/gguf/model-Q4_K_M.gguf +0 -0
- hermes3-llama31/gguf/reference.json +21 -0
- hermes3-llama31/inputs.safetensors +3 -0
- hermes3-llama31/package/config.json +35 -0
- hermes3-llama31/package/generation_config.json +6 -0
- hermes3-llama31/package/model.safetensors +3 -0
- hermes3-llama31/provenance.json +19 -0
- hermes3-llama31/reference/outputs.safetensors +3 -0
- hermes3-llama31/shrunk-config.json +35 -0
- hermes3-llama31/source-config.json +35 -0
- hermes3-llama31/tokenizer/metadata.json +24 -0
- hermes3-llama31/tokenizer/special_tokens_map.json +5 -0
- hermes3-llama31/tokenizer/tokenizer.model +3 -0
- hermes3-llama31/tokenizer/tokenizer_config.json +12 -0
- hermes3-llama31/tokenizer/vocabulary.json +130 -0
- hermes3-llama31/validation.json +74 -0
- livekit-turn-detector/case.json +58 -0
- livekit-turn-detector/config-diff.json +37 -0
- livekit-turn-detector/gguf/metadata.json +96 -0
- livekit-turn-detector/gguf/model-Q4_K_M.gguf +3 -0
- livekit-turn-detector/gguf/reference.json +21 -0
- livekit-turn-detector/inputs.safetensors +3 -0
- livekit-turn-detector/package/config.json +32 -0
.gitattributes
CHANGED
|
@@ -58,3 +58,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
livekit-turn-detector/gguf/model-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
minicpm5/gguf/model-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
tinyllama-chat/gguf/model-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
|
GGUF_Q4_K_M_ADDED_FOUR_REPORT.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cases": [
|
| 3 |
+
{
|
| 4 |
+
"case": "deepseek-coder",
|
| 5 |
+
"model_sha256": "be8a340c6563b75da1aba85732f68764cb84c4a36390e8060296940aeb1b1b05",
|
| 6 |
+
"model_size_bytes": 22784,
|
| 7 |
+
"status": "passed"
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"case": "hermes3-llama31",
|
| 11 |
+
"model_sha256": "e1b814e388a2f8611d6422f22d4426b14a6d9e7e6e69789d7f427ba5c525b8ef",
|
| 12 |
+
"model_size_bytes": 90688,
|
| 13 |
+
"status": "passed"
|
| 14 |
+
},
|
| 15 |
+
{
|
| 16 |
+
"case": "livekit-turn-detector",
|
| 17 |
+
"model_sha256": "14e5fc74d73f3723a80e449dbc58c5ee60a064f08d026e3743beef64d89c806f",
|
| 18 |
+
"model_size_bytes": 211456,
|
| 19 |
+
"status": "passed"
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
"case": "minicpm5",
|
| 23 |
+
"model_sha256": "b9e0e01198ac04fc0239b602007cd4fec8fb93a0e334361c2af92227e4ff8981",
|
| 24 |
+
"model_size_bytes": 966528,
|
| 25 |
+
"status": "passed"
|
| 26 |
+
}
|
| 27 |
+
],
|
| 28 |
+
"llama_cpp_revision": "40b740ad05c531b9d57aca6698c3ed553a9e784c",
|
| 29 |
+
"requested_quantization": "Q4_K_M",
|
| 30 |
+
"schema_version": 1
|
| 31 |
+
}
|
GGUF_Q4_K_M_REPORT.json
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cases": [
|
| 3 |
+
{
|
| 4 |
+
"case": "tinyllama-chat",
|
| 5 |
+
"model_sha256": "32bfddeebc736370580c83983cf01380a157c215600a5f1faef2bbb2ffdf9bed",
|
| 6 |
+
"model_size_bytes": 260160,
|
| 7 |
+
"status": "passed"
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"case": "smollm2-instruct",
|
| 11 |
+
"model_sha256": "8224570d8c01310dc10083e28da8b077503469fe06d701f44b047011065fda43",
|
| 12 |
+
"model_size_bytes": 87936,
|
| 13 |
+
"status": "passed"
|
| 14 |
+
},
|
| 15 |
+
{
|
| 16 |
+
"case": "mobilellama-chat",
|
| 17 |
+
"model_sha256": "c5f51217248a246dccd38037b1f1e3d9f1c0fce977a369fd8a848630259719f9",
|
| 18 |
+
"model_size_bytes": 22816,
|
| 19 |
+
"status": "passed"
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
"case": "deepseek-coder",
|
| 23 |
+
"model_sha256": "8557d99db9de6b588628f8fead576ff85b26edbdc8df176050b59a28acf720f3",
|
| 24 |
+
"model_size_bytes": 22784,
|
| 25 |
+
"status": "passed"
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"case": "hermes3-llama31",
|
| 29 |
+
"model_sha256": "6b53e66bd258178cb1ceb6f89ad36a3b909e912c430ec5ccac06a476d6a652e4",
|
| 30 |
+
"model_size_bytes": 90688,
|
| 31 |
+
"status": "passed"
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"case": "livekit-turn-detector",
|
| 35 |
+
"model_sha256": "29643572ec796cace456931e906b8357827a78df07007a937632bc72c4b20d0e",
|
| 36 |
+
"model_size_bytes": 211456,
|
| 37 |
+
"status": "passed"
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"case": "minicpm5",
|
| 41 |
+
"model_sha256": "b9e0e01198ac04fc0239b602007cd4fec8fb93a0e334361c2af92227e4ff8981",
|
| 42 |
+
"model_size_bytes": 966528,
|
| 43 |
+
"status": "passed"
|
| 44 |
+
}
|
| 45 |
+
],
|
| 46 |
+
"llama_cpp_revision": "40b740ad05c531b9d57aca6698c3ed553a9e784c",
|
| 47 |
+
"requested_quantization": "Q4_K_M",
|
| 48 |
+
"schema_version": 1
|
| 49 |
+
}
|
README.md
ADDED
|
@@ -0,0 +1,315 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Real-Config-Derived Tiny Llama Models
|
| 3 |
+
license: other
|
| 4 |
+
size_categories:
|
| 5 |
+
- n<1K
|
| 6 |
+
tags:
|
| 7 |
+
- llama
|
| 8 |
+
- transformers
|
| 9 |
+
- safetensors
|
| 10 |
+
- gguf
|
| 11 |
+
- synthetic
|
| 12 |
+
- conformance
|
| 13 |
+
- compatibility-testing
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# Real-Config-Derived Tiny Llama Models
|
| 17 |
+
|
| 18 |
+
This dataset contains small, deterministic Llama-compatible models derived from `config.json` files of real models published on the Hugging Face Hub.
|
| 19 |
+
|
| 20 |
+
The models are intended for model-format, loader, inference, compatibility, and conformance testing. They are not intended for language-model quality evaluation.
|
| 21 |
+
|
| 22 |
+
Each case preserves an observed real-world configuration as much as possible while reducing model geometry enough to make the resulting model inexpensive to store and execute.
|
| 23 |
+
|
| 24 |
+
The current collection contains seven Llama-family cases in both:
|
| 25 |
+
|
| 26 |
+
* unquantized F32 Safetensors format;
|
| 27 |
+
* GGUF `Q4_K_M` format.
|
| 28 |
+
|
| 29 |
+
All model weights in this dataset are synthetic and deterministically generated. No original model weights are included.
|
| 30 |
+
|
| 31 |
+
## Why this dataset exists
|
| 32 |
+
|
| 33 |
+
Small hand-written test models are useful, but they tend to exercise only a small number of clean, canonical configuration patterns.
|
| 34 |
+
|
| 35 |
+
Real model repositories contain considerably more variation, including:
|
| 36 |
+
|
| 37 |
+
* MHA and GQA configurations;
|
| 38 |
+
* different query/KV head ratios;
|
| 39 |
+
* explicit versus implicit `head_dim`;
|
| 40 |
+
* tied and untied embeddings;
|
| 41 |
+
* different special-token layouts;
|
| 42 |
+
* multiple EOS token IDs;
|
| 43 |
+
* older and newer RoPE scaling representations;
|
| 44 |
+
* configuration files produced by different Transformers versions.
|
| 45 |
+
|
| 46 |
+
This dataset retains such real configuration patterns while replacing the original large geometry and weights with small deterministic equivalents.
|
| 47 |
+
|
| 48 |
+
The goal is to provide compact test inputs that remain representative of model configurations that actually exist in the ecosystem.
|
| 49 |
+
|
| 50 |
+
## Cases
|
| 51 |
+
|
| 52 |
+
| Case | Source model | Parameters | F32 Safetensors | Notable configuration features |
|
| 53 |
+
| ----------------------- | ------------------------------------------ | ---------: | --------------: | ------------------------------------------ |
|
| 54 |
+
| `tinyllama-chat` | `TinyLlama/TinyLlama-1.1B-Chat-v1.0` | 303,744 | 1,217,160 B | GQA, query/KV ratio 8 |
|
| 55 |
+
| `smollm2-instruct` | `HuggingFaceTB/SmolLM2-135M-Instruct` | 46,320 | 187,328 B | GQA, query/KV ratio 3 |
|
| 56 |
+
| `mobilellama-chat` | `mtgv/MobileLLaMA-1.4B-Chat` | 9,296 | 39,288 B | MHA |
|
| 57 |
+
| `minicpm5` | `openbmb/MiniCPM5-1B` | 1,409,664 | 5,640,864 B | explicit `head_dim`, GQA, multiple EOS IDs |
|
| 58 |
+
| `deepseek-coder` | `deepseek-ai/deepseek-coder-6.7b-instruct` | 9,296 | 39,288 B | MHA, linear RoPE scaling |
|
| 59 |
+
| `hermes3-llama31` | `NousResearch/Hermes-3-Llama-3.1-8B` | 86,336 | 347,488 B | GQA, Llama-3 RoPE scaling |
|
| 60 |
+
| `livekit-turn-detector` | `livekit/turn-detector` | 132,336 | 531,400 B | explicit `head_dim`, GQA |
|
| 61 |
+
|
| 62 |
+
All seven source configurations passed the source-side validation performed with Transformers `4.55.0`, and all seven generated F32 models passed construction, prefill, cached decode, save, reload, and post-reload execution checks.
|
| 63 |
+
|
| 64 |
+
See `REPORT.md` for the complete shrink results and source revisions.
|
| 65 |
+
|
| 66 |
+
## Directory structure
|
| 67 |
+
|
| 68 |
+
Each case is self-contained:
|
| 69 |
+
|
| 70 |
+
```text
|
| 71 |
+
<case>/
|
| 72 |
+
├── case.json
|
| 73 |
+
├── provenance.json
|
| 74 |
+
├── source-config.json
|
| 75 |
+
├── shrunk-config.json
|
| 76 |
+
├── config-diff.json
|
| 77 |
+
├── validation.json
|
| 78 |
+
├── inputs.safetensors
|
| 79 |
+
│
|
| 80 |
+
├── package/
|
| 81 |
+
│ ├── config.json
|
| 82 |
+
│ ├── generation_config.json
|
| 83 |
+
│ └── model.safetensors
|
| 84 |
+
│
|
| 85 |
+
├── tokenizer/
|
| 86 |
+
│ ├── tokenizer.model
|
| 87 |
+
│ ├── tokenizer_config.json
|
| 88 |
+
│ ├── special_tokens_map.json
|
| 89 |
+
│ ├── vocabulary.json
|
| 90 |
+
│ └── metadata.json
|
| 91 |
+
│
|
| 92 |
+
├── reference/
|
| 93 |
+
│ └── outputs.safetensors
|
| 94 |
+
│
|
| 95 |
+
└── gguf/
|
| 96 |
+
├── model-Q4_K_M.gguf
|
| 97 |
+
├── metadata.json
|
| 98 |
+
└── reference.json
|
| 99 |
+
```
|
| 100 |
+
|
| 101 |
+
The top-level `manifest.json` provides machine-readable paths, hashes, and sizes for the available model packages and auxiliary tokenizers.
|
| 102 |
+
|
| 103 |
+
## Real configurations, synthetic models
|
| 104 |
+
|
| 105 |
+
A case has four distinct origins:
|
| 106 |
+
|
| 107 |
+
### Source configuration
|
| 108 |
+
|
| 109 |
+
`source-config.json` is the configuration obtained from the source Hugging Face repository at the immutable revision recorded in `provenance.json`.
|
| 110 |
+
|
| 111 |
+
Its SHA-256 is recorded for traceability.
|
| 112 |
+
|
| 113 |
+
### Shrunk configuration
|
| 114 |
+
|
| 115 |
+
`shrunk-config.json` is produced by a preservation-first Llama shrinker.
|
| 116 |
+
|
| 117 |
+
Only geometry-dependent fields required to make the model small are changed. Unrelated configuration fields are retained rather than normalized into a newly generated configuration.
|
| 118 |
+
|
| 119 |
+
The transformations used for this collection include reductions of fields such as:
|
| 120 |
+
|
| 121 |
+
* `hidden_size`;
|
| 122 |
+
* `intermediate_size`;
|
| 123 |
+
* `num_hidden_layers`;
|
| 124 |
+
* `num_attention_heads`;
|
| 125 |
+
* `num_key_value_heads`;
|
| 126 |
+
* `vocab_size`;
|
| 127 |
+
* `max_position_embeddings`.
|
| 128 |
+
|
| 129 |
+
Out-of-range special-token IDs are remapped only when required by vocabulary reduction.
|
| 130 |
+
|
| 131 |
+
Observed fields such as explicit `head_dim`, bias settings, embedding-tying settings, and the original RoPE scaling object are preserved where possible.
|
| 132 |
+
|
| 133 |
+
`config-diff.json` records the transformation for each case.
|
| 134 |
+
|
| 135 |
+
### Synthetic weights
|
| 136 |
+
|
| 137 |
+
The model weights are not derived from the original model checkpoint.
|
| 138 |
+
|
| 139 |
+
They are deterministically generated using `tlfloat::LCG64`, with the seed and exact tlfloat revision recorded in `provenance.json`.
|
| 140 |
+
|
| 141 |
+
This makes the generated model reproducible without distributing or downloading the original model weights.
|
| 142 |
+
|
| 143 |
+
### Synthetic auxiliary tokenizer
|
| 144 |
+
|
| 145 |
+
The reduced models use a vocabulary of 128 token IDs, so the original source-model tokenizers are generally not compatible with them.
|
| 146 |
+
|
| 147 |
+
Selected cases therefore contain a deterministic synthetic auxiliary tokenizer.
|
| 148 |
+
|
| 149 |
+
These tokenizers:
|
| 150 |
+
|
| 151 |
+
* cover the reduced token-ID space;
|
| 152 |
+
* preserve the special-token semantics required by the corresponding shrunk configuration;
|
| 153 |
+
* are used consistently for the Hugging Face and GGUF representations;
|
| 154 |
+
* are not derived from the original source tokenizer;
|
| 155 |
+
* do not attempt to reproduce the linguistic tokenization behavior of the original model.
|
| 156 |
+
|
| 157 |
+
The tokenizer metadata explicitly records this distinction.
|
| 158 |
+
|
| 159 |
+
For numerical model testing, explicit token IDs remain the primary input. Text tokenization is secondary.
|
| 160 |
+
|
| 161 |
+
## Safetensors packages
|
| 162 |
+
|
| 163 |
+
The `package/` directory is a normal small Hugging Face model package containing the shrunk configuration and deterministic F32 weights.
|
| 164 |
+
|
| 165 |
+
Reference inputs are stored separately in `inputs.safetensors`, and expected F32 outputs are stored in `reference/outputs.safetensors`.
|
| 166 |
+
|
| 167 |
+
These reference artifacts were generated and validated with Transformers `4.55.0`.
|
| 168 |
+
|
| 169 |
+
Exact software versions and hashes are available in each case's `provenance.json`.
|
| 170 |
+
|
| 171 |
+
## GGUF Q4_K_M packages
|
| 172 |
+
|
| 173 |
+
Every case also contains:
|
| 174 |
+
|
| 175 |
+
```text
|
| 176 |
+
gguf/model-Q4_K_M.gguf
|
| 177 |
+
```
|
| 178 |
+
|
| 179 |
+
The GGUF files are derived from the same deterministic model weights as the Safetensors packages.
|
| 180 |
+
|
| 181 |
+
They were generated with upstream `ggml-org/llama.cpp` at commit:
|
| 182 |
+
|
| 183 |
+
```text
|
| 184 |
+
40b740ad05c531b9d57aca6698c3ed553a9e784c
|
| 185 |
+
```
|
| 186 |
+
|
| 187 |
+
The conversion pipeline is:
|
| 188 |
+
|
| 189 |
+
```text
|
| 190 |
+
F32 Safetensors model
|
| 191 |
+
↓
|
| 192 |
+
llama.cpp HF-to-GGUF converter
|
| 193 |
+
↓
|
| 194 |
+
temporary F16 GGUF
|
| 195 |
+
↓
|
| 196 |
+
llama-quantize Q4_K_M
|
| 197 |
+
↓
|
| 198 |
+
retained Q4_K_M GGUF
|
| 199 |
+
```
|
| 200 |
+
|
| 201 |
+
The temporary F16 GGUF is hashed for provenance but is not retained.
|
| 202 |
+
|
| 203 |
+
`Q4_K_M` is a lossy mixed quantization profile. In very small models, some tensor dimensions are too small for particular K-quant types, so the final GGUF may contain fallback tensor types.
|
| 204 |
+
|
| 205 |
+
For this reason, each `gguf/metadata.json` records the actual tensor-type histogram found in the resulting file.
|
| 206 |
+
|
| 207 |
+
For example, depending on the case, a `Q4_K_M` artifact may contain a mixture of:
|
| 208 |
+
|
| 209 |
+
* F16;
|
| 210 |
+
* F32;
|
| 211 |
+
* Q4_K;
|
| 212 |
+
* Q5_0;
|
| 213 |
+
* Q6_K;
|
| 214 |
+
* Q8_0.
|
| 215 |
+
|
| 216 |
+
The label `Q4_K_M` identifies the requested llama.cpp quantization profile; it does not imply that every tensor is stored as Q4_K.
|
| 217 |
+
|
| 218 |
+
## GGUF validation
|
| 219 |
+
|
| 220 |
+
GGUF numerical validation uses explicit token IDs rather than text tokenization.
|
| 221 |
+
|
| 222 |
+
For each case, the final GGUF was loaded through the same pinned llama.cpp revision and tested with deterministic:
|
| 223 |
+
|
| 224 |
+
* prefill;
|
| 225 |
+
* autoregressive decode;
|
| 226 |
+
* logit extraction;
|
| 227 |
+
* finite-value checks;
|
| 228 |
+
* repeated-execution checks.
|
| 229 |
+
|
| 230 |
+
The effective EOG token set was also checked over the complete reduced token-ID range and required to match the EOS semantics of the shrunk configuration.
|
| 231 |
+
|
| 232 |
+
`gguf/reference.json` contains the GGUF reference output, while `gguf/metadata.json` records:
|
| 233 |
+
|
| 234 |
+
* llama.cpp revision;
|
| 235 |
+
* source-package SHA-256;
|
| 236 |
+
* tokenizer SHA-256;
|
| 237 |
+
* converter and quantizer invocations;
|
| 238 |
+
* intermediate and final GGUF hashes;
|
| 239 |
+
* actual tensor-type histogram;
|
| 240 |
+
* effective EOG token IDs;
|
| 241 |
+
* direct-token validation results;
|
| 242 |
+
* observed numerical difference from the F32 reference.
|
| 243 |
+
|
| 244 |
+
Because `Q4_K_M` is lossy, GGUF logits are not expected to be numerically identical to the F32 Transformers reference.
|
| 245 |
+
|
| 246 |
+
The quantization differences are observations rather than exact-equivalence requirements.
|
| 247 |
+
|
| 248 |
+
## Reproducibility and provenance
|
| 249 |
+
|
| 250 |
+
Each case records enough information to trace:
|
| 251 |
+
|
| 252 |
+
```text
|
| 253 |
+
real Hugging Face config + immutable revision
|
| 254 |
+
↓
|
| 255 |
+
preservation-first shrink
|
| 256 |
+
↓
|
| 257 |
+
deterministic small config
|
| 258 |
+
↓
|
| 259 |
+
deterministic synthetic weights
|
| 260 |
+
↙ ↘
|
| 261 |
+
F32 Safetensors auxiliary tokenizer
|
| 262 |
+
\ /
|
| 263 |
+
\ /
|
| 264 |
+
GGUF conversion
|
| 265 |
+
↓
|
| 266 |
+
Q4_K_M GGUF
|
| 267 |
+
```
|
| 268 |
+
|
| 269 |
+
Relevant identities are represented by SHA-256 hashes rather than filenames or timestamps.
|
| 270 |
+
|
| 271 |
+
See:
|
| 272 |
+
|
| 273 |
+
* `provenance.json` for the source and F32-generation chain;
|
| 274 |
+
* `tokenizer/metadata.json` for tokenizer provenance;
|
| 275 |
+
* `gguf/metadata.json` for GGUF provenance;
|
| 276 |
+
* `manifest.json` for the dataset-wide package index.
|
| 277 |
+
|
| 278 |
+
## What this dataset is not
|
| 279 |
+
|
| 280 |
+
This is not:
|
| 281 |
+
|
| 282 |
+
* a collection of pretrained language models;
|
| 283 |
+
* a benchmark of model quality;
|
| 284 |
+
* a reproduction of the source models' weights;
|
| 285 |
+
* a reproduction of the source models' tokenizers;
|
| 286 |
+
* a recommendation to use these geometries for training;
|
| 287 |
+
* a comprehensive set of every possible Llama configuration.
|
| 288 |
+
|
| 289 |
+
The value of the dataset is the combination of real observed configuration patterns with small, deterministic, executable model artifacts.
|
| 290 |
+
|
| 291 |
+
## Source models and licensing
|
| 292 |
+
|
| 293 |
+
The source configurations originate from the Hugging Face repositories identified in each case's `provenance.json`.
|
| 294 |
+
|
| 295 |
+
The generated model weights and auxiliary tokenizer data are synthetic, but source configuration files and associated metadata may remain subject to terms applicable to their respective source repositories.
|
| 296 |
+
|
| 297 |
+
This repository therefore does not claim that a single upstream model license applies uniformly to every source-derived file.
|
| 298 |
+
|
| 299 |
+
Users should consult the corresponding source repository when redistributing or using source-derived metadata under conditions where its license is relevant.
|
| 300 |
+
|
| 301 |
+
No original source-model checkpoint weights are included.
|
| 302 |
+
|
| 303 |
+
## Reports
|
| 304 |
+
|
| 305 |
+
* `REPORT.md` — Llama shrinker results and source provenance summary.
|
| 306 |
+
* `GGUF_Q4_K_M_REPORT.json` — consolidated GGUF generation and validation report.
|
| 307 |
+
* `manifest.json` — machine-readable package inventory.
|
| 308 |
+
|
| 309 |
+
## Current scope
|
| 310 |
+
|
| 311 |
+
The current release contains seven Llama-compatible cases.
|
| 312 |
+
|
| 313 |
+
The collection is intentionally small: it favors configurations observed in real, relatively popular model repositories rather than generating large numbers of artificial configuration combinations.
|
| 314 |
+
|
| 315 |
+
Future collections may add other model families or additional observed configuration patterns while retaining the same principles of small artifacts, deterministic generation, explicit provenance, and format-native validation.
|
REPORT.md
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Llama shrinker v1 report
|
| 2 |
+
|
| 3 |
+
Transformers: `4.55.0`
|
| 4 |
+
Allowed modified fields: `bos_token_id, eos_token_id, hidden_size, intermediate_size, max_position_embeddings, num_attention_heads, num_hidden_layers, num_key_value_heads, pad_token_id, vocab_size`
|
| 5 |
+
|
| 6 |
+
| Repository | Source pre-check | Config / model class | Original geometry | Shrunk geometry | Modified fields | Parameters | Checkpoint bytes | Post-check |
|
| 7 |
+
| --- | --- | --- | --- | --- | --- | ---: | ---: | --- |
|
| 8 |
+
| `TinyLlama/TinyLlama-1.1B-Chat-v1.0` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":2048,"intermediate_size":5632,"num_hidden_layers":22,"num_attention_heads":32,"num_key_value_heads":4,"vocab_size":32000,"max_position_embeddings":2048}` | `{"hidden_size":128,"intermediate_size":256,"num_hidden_layers":2,"num_attention_heads":16,"num_key_value_heads":2,"vocab_size":128,"max_position_embeddings":128}` | `hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 303744 | 1217160 | POSTCHECK_OK |
|
| 9 |
+
| `HuggingFaceTB/SmolLM2-135M-Instruct` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":576,"intermediate_size":1536,"num_hidden_layers":30,"num_attention_heads":9,"num_key_value_heads":3,"vocab_size":49152,"max_position_embeddings":8192}` | `{"hidden_size":48,"intermediate_size":96,"num_hidden_layers":2,"num_attention_heads":6,"num_key_value_heads":2,"vocab_size":128,"max_position_embeddings":128}` | `hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 46320 | 187328 | POSTCHECK_OK |
|
| 10 |
+
| `mtgv/MobileLLaMA-1.4B-Chat` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":2048,"intermediate_size":5632,"num_hidden_layers":24,"num_attention_heads":16,"num_key_value_heads":16,"vocab_size":32000,"max_position_embeddings":2048}` | `{"hidden_size":16,"intermediate_size":32,"num_hidden_layers":2,"num_attention_heads":2,"num_key_value_heads":2,"vocab_size":128,"max_position_embeddings":128}` | `hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 9296 | 39288 | POSTCHECK_OK |
|
| 11 |
+
| `openbmb/MiniCPM5-1B` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":1536,"intermediate_size":4608,"num_hidden_layers":24,"num_attention_heads":16,"num_key_value_heads":2,"head_dim":128,"vocab_size":130560,"max_position_embeddings":131072}` | `{"hidden_size":128,"intermediate_size":256,"num_hidden_layers":2,"num_attention_heads":16,"num_key_value_heads":2,"head_dim":128,"vocab_size":128,"max_position_embeddings":128}` | `eos_token_id,hidden_size,intermediate_size,max_position_embeddings,num_hidden_layers,vocab_size` | 1409664 | 5640864 | POSTCHECK_OK |
|
| 12 |
+
| `deepseek-ai/deepseek-coder-6.7b-instruct` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":4096,"intermediate_size":11008,"num_hidden_layers":32,"num_attention_heads":32,"num_key_value_heads":32,"vocab_size":32256,"max_position_embeddings":16384}` | `{"hidden_size":16,"intermediate_size":32,"num_hidden_layers":2,"num_attention_heads":2,"num_key_value_heads":2,"vocab_size":128,"max_position_embeddings":128}` | `bos_token_id,eos_token_id,hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 9296 | 39288 | POSTCHECK_OK |
|
| 13 |
+
| `NousResearch/Hermes-3-Llama-3.1-8B` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":4096,"intermediate_size":14336,"num_hidden_layers":32,"num_attention_heads":32,"num_key_value_heads":8,"vocab_size":128256,"max_position_embeddings":131072}` | `{"hidden_size":64,"intermediate_size":128,"num_hidden_layers":2,"num_attention_heads":8,"num_key_value_heads":2,"vocab_size":128,"max_position_embeddings":128}` | `bos_token_id,eos_token_id,hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 86336 | 347488 | POSTCHECK_OK |
|
| 14 |
+
| `livekit/turn-detector` | SOURCE_OK | LlamaConfig / LlamaForCausalLM | `{"hidden_size":576,"intermediate_size":1536,"num_hidden_layers":30,"num_attention_heads":9,"num_key_value_heads":3,"head_dim":64,"vocab_size":49154,"max_position_embeddings":8192}` | `{"hidden_size":48,"intermediate_size":96,"num_hidden_layers":2,"num_attention_heads":6,"num_key_value_heads":2,"head_dim":64,"vocab_size":128,"max_position_embeddings":128}` | `hidden_size,intermediate_size,max_position_embeddings,num_attention_heads,num_hidden_layers,num_key_value_heads,vocab_size` | 132336 | 531400 | POSTCHECK_OK |
|
| 15 |
+
|
| 16 |
+
## Per-source provenance
|
| 17 |
+
|
| 18 |
+
- `TinyLlama/TinyLlama-1.1B-Chat-v1.0`: revision `fe8a4ea1ffedaf415f4da2f062534de366a451e6`, source SHA-256 `486bedda3a6988332e60d9638a09ca4b260d34ebcf1b19e22cf3b140b63d8fe9`, rules `gqa-ratio-8-preserved`.
|
| 19 |
+
|
| 20 |
+
- `HuggingFaceTB/SmolLM2-135M-Instruct`: revision `12fd25f77366fa6b3b4b768ec3050bf629380bac`, source SHA-256 `8eb740e8bbe4cff95ea7b4588d17a2432deb16e8075bc5828ff7ba9be94d982a`, rules `gqa-ratio-3-preserved`.
|
| 21 |
+
|
| 22 |
+
- `mtgv/MobileLLaMA-1.4B-Chat`: revision `7073081bd6b02d97ac4f9c7e8167fd3c28483c1c`, source SHA-256 `1e0f3e66ba7c3824ae2a2352df692c50d0e1b80df3b1fd1338670084de905aaf`, rules `mha-preserved`.
|
| 23 |
+
|
| 24 |
+
- `openbmb/MiniCPM5-1B`: revision `4e9de7a0778dc1c362e983e6858f0e77542cbdca`, source SHA-256 `6a6509b646cb3169616c5ffc3196e7ccaf9d4d6bc17b266581d241a31c217714`, rules `gqa-ratio-8-preserved, out-of-range-special-tokens-remapped, explicit-head-dim-preserved`.
|
| 25 |
+
|
| 26 |
+
- `deepseek-ai/deepseek-coder-6.7b-instruct`: revision `e5d64addd26a6a1db0f9b863abf6ee3141936807`, source SHA-256 `a3a2ce1a623b367d3799fb0c1e1ba12d5b8fa51a8e7abba14d10dd3739cd73e4`, rules `mha-preserved, out-of-range-special-tokens-remapped, rope-scaling-preserved`.
|
| 27 |
+
|
| 28 |
+
- `NousResearch/Hermes-3-Llama-3.1-8B`: revision `896ea440e5a9e6070e3d8a2774daf2b481ab425b`, source SHA-256 `7e189a0da074da6a878d5a276336caf3d62c57138f461bc6239402639f1bc6f5`, rules `gqa-ratio-4-preserved, out-of-range-special-tokens-remapped, rope-scaling-preserved`.
|
| 29 |
+
|
| 30 |
+
- `livekit/turn-detector`: revision `fba34c38ad5d30a63ebb83a9e6bf271cf4c91d67`, source SHA-256 `e8feb37fa6f974232742d4e223e40928ccc8dabe1906ce97090e14b5753dbcc1`, rules `gqa-ratio-3-preserved, explicit-head-dim-preserved`.
|
| 31 |
+
|
| 32 |
+
## Global findings
|
| 33 |
+
|
| 34 |
+
One repository-independent shrinker handled 7/7 source configs that passed pre-check.
|
| 35 |
+
|
| 36 |
+
Dependency rules: query heads remain divisible by KV heads; MHA equality is preserved; GQA keeps its observed query/KV ratio and at least two KV heads; hidden width is eight values per reduced query head; intermediate width is twice hidden width; context is reduced to 128 positions; only out-of-range special IDs are remapped while preserving representation and equality; explicit `head_dim`, both observed `rope_scaling` forms, and semantic bias/tie fields are preserved.
|
| 37 |
+
|
| 38 |
+
Special treatment was required only for explicit `head_dim` preservation and deterministic special-token remapping. Both the linear and Llama-3 RoPE objects execute without dependent changes after context is reduced to 128. Transformers emits a non-fatal consistency warning for the preserved Llama-3 `original_max_position_embeddings=8192`, which is recorded as a validation observation.
|
| 39 |
+
|
| 40 |
+
Remaining v0 limitation: MQA and unobserved special-token field names are intentionally not generalized from this seven-config corpus.
|
deepseek-coder/case.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architecture": "llama",
|
| 3 |
+
"artifact_id": "deepseek-coder",
|
| 4 |
+
"auxiliary_tokenizer": {
|
| 5 |
+
"metadata": "tokenizer/metadata.json",
|
| 6 |
+
"path": "tokenizer"
|
| 7 |
+
},
|
| 8 |
+
"comparison_policy": {
|
| 9 |
+
"cached_vs_cache_free": {
|
| 10 |
+
"atol": 1e-06,
|
| 11 |
+
"rtol": 1e-05
|
| 12 |
+
},
|
| 13 |
+
"primary_value": "logits",
|
| 14 |
+
"reload": {
|
| 15 |
+
"atol": 1e-07,
|
| 16 |
+
"rtol": 1e-06
|
| 17 |
+
},
|
| 18 |
+
"require_finite": true
|
| 19 |
+
},
|
| 20 |
+
"format": {
|
| 21 |
+
"container": "huggingface-safetensors",
|
| 22 |
+
"dtype": "float32",
|
| 23 |
+
"layout": "single-file",
|
| 24 |
+
"package_path": "package"
|
| 25 |
+
},
|
| 26 |
+
"packages": {
|
| 27 |
+
"gguf_q4_k_m": {
|
| 28 |
+
"metadata": "gguf/metadata.json",
|
| 29 |
+
"model": "gguf/model-Q4_K_M.gguf",
|
| 30 |
+
"reference": "gguf/reference.json"
|
| 31 |
+
},
|
| 32 |
+
"safetensors_f32": {
|
| 33 |
+
"config": "package/config.json",
|
| 34 |
+
"model": "package/model.safetensors",
|
| 35 |
+
"sha256": "cba3bb2298077f3bdeee9da78a4d0ae7586c3a3964e7a40eb2f9e5dc2631c4cc",
|
| 36 |
+
"size_bytes": 39288
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"reference_runtime": {
|
| 40 |
+
"config_class": "LlamaConfig",
|
| 41 |
+
"implementation": "huggingface-transformers",
|
| 42 |
+
"model_class": "LlamaForCausalLM",
|
| 43 |
+
"trust_remote_code": false,
|
| 44 |
+
"version": "4.55.0"
|
| 45 |
+
},
|
| 46 |
+
"schema_version": 1,
|
| 47 |
+
"semantic_features": {
|
| 48 |
+
"explicit_head_dim": false,
|
| 49 |
+
"non_null_rope_scaling": true,
|
| 50 |
+
"query_width_matches_hidden_size": true
|
| 51 |
+
},
|
| 52 |
+
"source": {
|
| 53 |
+
"config_sha256": "a3a2ce1a623b367d3799fb0c1e1ba12d5b8fa51a8e7abba14d10dd3739cd73e4",
|
| 54 |
+
"repository": "deepseek-ai/deepseek-coder-6.7b-instruct",
|
| 55 |
+
"revision": "e5d64addd26a6a1db0f9b863abf6ee3141936807"
|
| 56 |
+
},
|
| 57 |
+
"validation": "POSTCHECK_OK"
|
| 58 |
+
}
|
deepseek-coder/config-diff.json
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"path": "/bos_token_id",
|
| 4 |
+
"source": 32013,
|
| 5 |
+
"shrunk": 0
|
| 6 |
+
},
|
| 7 |
+
{
|
| 8 |
+
"path": "/eos_token_id",
|
| 9 |
+
"source": 32021,
|
| 10 |
+
"shrunk": 1
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"path": "/hidden_size",
|
| 14 |
+
"source": 4096,
|
| 15 |
+
"shrunk": 16
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"path": "/intermediate_size",
|
| 19 |
+
"source": 11008,
|
| 20 |
+
"shrunk": 32
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"path": "/max_position_embeddings",
|
| 24 |
+
"source": 16384,
|
| 25 |
+
"shrunk": 128
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"path": "/num_attention_heads",
|
| 29 |
+
"source": 32,
|
| 30 |
+
"shrunk": 2
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"path": "/num_hidden_layers",
|
| 34 |
+
"source": 32,
|
| 35 |
+
"shrunk": 2
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"path": "/num_key_value_heads",
|
| 39 |
+
"source": 32,
|
| 40 |
+
"shrunk": 2
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"path": "/vocab_size",
|
| 44 |
+
"source": 32256,
|
| 45 |
+
"shrunk": 128
|
| 46 |
+
}
|
| 47 |
+
]
|
deepseek-coder/gguf/metadata.json
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"commands": {
|
| 3 |
+
"convert": [
|
| 4 |
+
"PYTHON",
|
| 5 |
+
"LLAMA_CPP/convert_hf_to_gguf.py",
|
| 6 |
+
"STAGING",
|
| 7 |
+
"--outfile",
|
| 8 |
+
"INTERMEDIATE",
|
| 9 |
+
"--outtype",
|
| 10 |
+
"f16"
|
| 11 |
+
],
|
| 12 |
+
"quantize": [
|
| 13 |
+
"LLAMA_QUANTIZE",
|
| 14 |
+
"--output-tensor-type",
|
| 15 |
+
"f16",
|
| 16 |
+
"--token-embedding-type",
|
| 17 |
+
"f16",
|
| 18 |
+
"INTERMEDIATE",
|
| 19 |
+
"OUTPUT",
|
| 20 |
+
"Q4_K_M"
|
| 21 |
+
]
|
| 22 |
+
},
|
| 23 |
+
"comparison_to_f32_reference": {
|
| 24 |
+
"decode": {
|
| 25 |
+
"max_abs_error": 6.486475467681885e-05,
|
| 26 |
+
"mean_abs_error": 2.3104150386643596e-05,
|
| 27 |
+
"top1_agreement": true,
|
| 28 |
+
"top5_overlap": 5
|
| 29 |
+
},
|
| 30 |
+
"prefill": {
|
| 31 |
+
"max_abs_error": 7.614446803927422e-05,
|
| 32 |
+
"mean_abs_error": 2.0526116713881493e-05,
|
| 33 |
+
"top1_agreement": true,
|
| 34 |
+
"top5_overlap": 5
|
| 35 |
+
}
|
| 36 |
+
},
|
| 37 |
+
"converter": "convert_hf_to_gguf.py",
|
| 38 |
+
"converter_log_sha256": "5029f6e7c0af519f829e53b9db930b09f62ea2444d3d267ca839d7202ed6c3eb",
|
| 39 |
+
"direct_token_validation": {
|
| 40 |
+
"all_logits_finite": true,
|
| 41 |
+
"decode_token_id": 13,
|
| 42 |
+
"exactly_reproducible": true,
|
| 43 |
+
"prefill_token_ids": [
|
| 44 |
+
3,
|
| 45 |
+
5,
|
| 46 |
+
7,
|
| 47 |
+
11
|
| 48 |
+
]
|
| 49 |
+
},
|
| 50 |
+
"effective_eog_token_ids": [
|
| 51 |
+
1
|
| 52 |
+
],
|
| 53 |
+
"eog_validation_range": [
|
| 54 |
+
0,
|
| 55 |
+
127
|
| 56 |
+
],
|
| 57 |
+
"format": "GGUF",
|
| 58 |
+
"intermediate_f16_sha256": "58fe175f8292f78728181fe74107e7bdc107748900cc5e3e7cc9e628daaad4ea",
|
| 59 |
+
"intermediate_retained": false,
|
| 60 |
+
"llama_cpp_revision": "40b740ad05c531b9d57aca6698c3ed553a9e784c",
|
| 61 |
+
"model_sha256": "8557d99db9de6b588628f8fead576ff85b26edbdc8df176050b59a28acf720f3",
|
| 62 |
+
"model_size_bytes": 22784,
|
| 63 |
+
"notes": "Q4_K_M is lossy; numerical differences are recorded as observations, not pass thresholds.",
|
| 64 |
+
"quantization_requested": "Q4_K_M",
|
| 65 |
+
"quantizer": "llama-quantize",
|
| 66 |
+
"quantizer_fallback_observed": true,
|
| 67 |
+
"quantizer_log_sha256": "91680988560477d86ec6ecc8b7e3b7b6fd21236183dd3c918afb9f4c69e07f85",
|
| 68 |
+
"schema_version": 1,
|
| 69 |
+
"source_auxiliary_tokenizer_sha256": "0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee",
|
| 70 |
+
"source_package_model_sha256": "cba3bb2298077f3bdeee9da78a4d0ae7586c3a3964e7a40eb2f9e5dc2631c4cc",
|
| 71 |
+
"staging_special_token_normalization": {
|
| 72 |
+
"applied": false,
|
| 73 |
+
"gguf_primary_eos_token_id": 1,
|
| 74 |
+
"reason": "GGUF exposes one primary EOS metadata field; all source EOS IDs remain documented by tokenizer metadata.",
|
| 75 |
+
"source_eos_token_id": 1
|
| 76 |
+
},
|
| 77 |
+
"tensor_type_histogram": {
|
| 78 |
+
"F16": 14,
|
| 79 |
+
"F32": 5,
|
| 80 |
+
"Q5_0": 1,
|
| 81 |
+
"Q8_0": 1
|
| 82 |
+
},
|
| 83 |
+
"tiny_dimension_tensor_overrides": {
|
| 84 |
+
"output.weight": "F16",
|
| 85 |
+
"reason": "These tensors can be too narrow for the Q8_0 types selected by the Q4_K_M mixture.",
|
| 86 |
+
"token_embd.weight": "F16"
|
| 87 |
+
},
|
| 88 |
+
"tokenizer_smoke": {
|
| 89 |
+
"hf_native_ids_equal": true,
|
| 90 |
+
"text": "a",
|
| 91 |
+
"token_ids": [
|
| 92 |
+
126,
|
| 93 |
+
124
|
| 94 |
+
]
|
| 95 |
+
}
|
| 96 |
+
}
|
deepseek-coder/gguf/model-Q4_K_M.gguf
ADDED
|
Binary file (22.8 kB). View file
|
|
|
deepseek-coder/gguf/reference.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"model_type": "llama",
|
| 4 |
+
"file_type": 15,
|
| 5 |
+
"tokenizer_model": "llama",
|
| 6 |
+
"vocab_size": 128,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"pad_token_id": -1,
|
| 10 |
+
"declared_bos_token_id": 0,
|
| 11 |
+
"declared_eos_token_id": 1,
|
| 12 |
+
"declared_pad_token_id": -1,
|
| 13 |
+
"eog_token_ids": [1],
|
| 14 |
+
"prefill_token_ids": [3,5,7,11],
|
| 15 |
+
"decode_token_id": 13,
|
| 16 |
+
"smoke_text": "a",
|
| 17 |
+
"smoke_token_ids": [126,124],
|
| 18 |
+
"all_logits_finite": true,
|
| 19 |
+
"prefill_last_logits": [0.0804776177,0.050990507,-0.047145199,-0.0296839438,0.024589939,-0.0976159945,0.0953064114,0.0776557848,-0.0127602071,0.0505206324,-0.0931572989,-0.0105060358,-0.0499786362,-0.00437465776,-0.0273610186,-0.0806470811,0.0947221816,0.0655309558,0.0300234016,-0.0997114554,-0.071042873,0.0178382322,-0.0651391447,0.0944337472,-0.0607835948,-0.100462899,0.0208385326,0.0885870904,0.0423921384,0.0122393016,-0.124106899,0.132509574,0.0756797418,-0.0954602733,0.0470732413,-0.0372419581,0.0169914402,-0.0120248748,-0.0854612663,-0.0126948338,0.0457908176,-0.0367843919,-0.0104824118,-0.10317982,-0.0565918386,0.0215061232,0.0205506533,0.0925647616,0.125231609,0.0859998912,-0.0601814613,0.00790851936,-0.0145831201,-0.00607665116,-0.0156599376,0.0132267792,0.0231828727,0.0171829592,0.0192843024,-0.0305071622,-0.0720497146,-0.0200255997,-0.0564342067,-0.0145750754,-0.0693934262,-0.0415456668,-0.0985182151,-0.0463196859,0.0160763841,-0.032519538,0.0501827523,0.0391519517,-0.0593523495,0.041016493,-0.0123989983,0.114316911,-0.0081050517,-0.013001374,0.0805320069,0.012509129,-0.0693908259,-0.0130453892,-0.0466823131,0.0578764379,0.0168398358,0.0536444113,-0.101287171,-0.0225926526,-0.0730922818,0.014947271,0.0367594361,0.0322937444,0.0444535576,-0.0544717796,0.0742822215,0.0167700183,-0.0107664131,0.0541727021,-0.0546422191,0.0634326413,0.00241968455,-0.0225936472,-0.036322251,-0.0327481478,0.120528094,0.0243949778,0.0285644103,-0.0178457964,0.0211905837,-0.0126141412,-0.0589847043,0.0164953712,0.0169796981,-0.0025812434,-0.0337066539,0.0365490355,-7.4907206e-05,-0.0457871482,0.0541579947,0.100902401,-0.0544508472,0.00696848659,-0.0338569805,-0.055723317,-0.0929735452,-0.0266074613,-0.0058335755,-0.0278450809],
|
| 20 |
+
"decode_logits": [0.0392578728,0.00496754097,-0.00843134522,-0.0548535697,-0.0540529527,-0.0983000919,-0.0183867533,-0.031128265,0.0357725844,-0.100170039,-0.0308085289,0.0151854865,0.0852568224,0.0314441361,0.0397567376,-0.00998981856,0.0531556383,-0.00594878849,0.0315914825,0.00264483457,-0.0280332565,-0.0100634946,0.036051102,-0.0593013167,-0.0440843925,-0.0144481547,-0.0805145651,0.0942398757,0.0532309078,-0.0397380814,-0.0521098413,0.0815597773,0.0677062124,-0.00898297783,0.0492810011,-0.0777892321,0.0187144876,0.00845821202,0.0515651777,0.0151565848,-0.014511236,-0.0480086096,-0.0501973629,0.0162530113,-0.0931950435,-0.0490497872,0.0608823188,-0.0255502351,0.00131554669,-0.00202353252,0.0231558233,-0.0109166848,-0.000116673356,-0.0413161702,0.0628425106,-0.0433886796,0.0710559264,-0.0260448903,-0.000890853291,-0.00447492814,-0.055997666,0.140653431,0.076936245,0.0150072379,-0.0672230199,0.0154967662,0.0706945434,-0.0208442211,0.0391199812,0.0674774274,-0.0650374368,0.0814273283,0.0532460213,-0.00137442758,-0.0491142012,0.0090700537,-0.00416620169,0.0408929847,-0.02375843,0.0257755574,-0.0418396667,0.00597975962,-0.0755510777,-0.0124426885,-0.0190046374,0.0364807174,0.0172516704,-0.0420099273,-0.0247528367,-0.0342988968,-0.0422521494,-0.0302511416,-0.0739510208,-0.0631983355,-0.00550435437,-0.0699428841,0.0739962459,-0.0493020415,0.0477355197,0.0171863437,-0.00883770827,-0.00670761801,-0.0486388467,-0.0281473789,-0.0190905575,0.0105521493,0.0295086447,0.00976369996,0.0173547808,0.0433472395,0.0135215661,0.0106512224,-0.0770340711,0.0830015987,0.0369128585,-0.13149713,0.0493771918,0.0255489107,0.0565564968,-0.00259162183,0.0013429831,-0.0835407078,0.0219996776,-0.0496292599,-0.0685657486,0.0625146627,-0.0146035366,-0.0409057364]
|
| 21 |
+
}
|
deepseek-coder/inputs.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b49b7cca71e662452eb97b184578af4f59e9a93bc541d621986836cd7e2c395
|
| 3 |
+
size 176
|
deepseek-coder/package/config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "/3fs-jd/prod/deepseek/shared/zhuqihao/public_model/deepseek-coder-7b-instruct2",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"hidden_act": "silu",
|
| 10 |
+
"hidden_size": 16,
|
| 11 |
+
"initializer_range": 0.02,
|
| 12 |
+
"intermediate_size": 32,
|
| 13 |
+
"max_position_embeddings": 128,
|
| 14 |
+
"model_type": "llama",
|
| 15 |
+
"num_attention_heads": 2,
|
| 16 |
+
"num_hidden_layers": 2,
|
| 17 |
+
"num_key_value_heads": 2,
|
| 18 |
+
"pretraining_tp": 1,
|
| 19 |
+
"rms_norm_eps": 1e-06,
|
| 20 |
+
"rope_scaling": {
|
| 21 |
+
"factor": 4.0,
|
| 22 |
+
"type": "linear"
|
| 23 |
+
},
|
| 24 |
+
"rope_theta": 100000,
|
| 25 |
+
"tie_word_embeddings": false,
|
| 26 |
+
"torch_dtype": "bfloat16",
|
| 27 |
+
"transformers_version": "4.34.1",
|
| 28 |
+
"use_cache": true,
|
| 29 |
+
"vocab_size": 128
|
| 30 |
+
}
|
deepseek-coder/package/generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 0,
|
| 4 |
+
"eos_token_id": 1,
|
| 5 |
+
"transformers_version": "4.55.0"
|
| 6 |
+
}
|
deepseek-coder/package/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cba3bb2298077f3bdeee9da78a4d0ae7586c3a3964e7a40eb2f9e5dc2631c4cc
|
| 3 |
+
size 39288
|
deepseek-coder/provenance.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"repository": "deepseek-ai/deepseek-coder-6.7b-instruct",
|
| 4 |
+
"revision": "e5d64addd26a6a1db0f9b863abf6ee3141936807",
|
| 5 |
+
"source_config_sha256": "a3a2ce1a623b367d3799fb0c1e1ba12d5b8fa51a8e7abba14d10dd3739cd73e4",
|
| 6 |
+
"shrinker_version": "llama-preservation-v1",
|
| 7 |
+
"conf_tool_version": "0.2.0",
|
| 8 |
+
"seed": 4850181387854826561,
|
| 9 |
+
"rng": "tlfloat::LCG64 from testerutil.hpp",
|
| 10 |
+
"tlfloat_revision": "936f0b093f0a79c0534fe5185ff80b6a649c576c",
|
| 11 |
+
"generated_config_sha256": "1385d0151ad8bfe96845e0301e85e01cdab09ba16c11348e838754a2fb9505c9",
|
| 12 |
+
"generated_checkpoint_sha256": "cba3bb2298077f3bdeee9da78a4d0ae7586c3a3964e7a40eb2f9e5dc2631c4cc",
|
| 13 |
+
"python": "3.12.13",
|
| 14 |
+
"torch": "2.8.0+cu128",
|
| 15 |
+
"transformers": "4.55.0",
|
| 16 |
+
"accelerate": "1.10.0",
|
| 17 |
+
"safetensors": "0.6.2",
|
| 18 |
+
"huggingface_hub": "0.34.4"
|
| 19 |
+
}
|
deepseek-coder/reference/outputs.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4ceef3c5d1268f4fea0922628dab26f78f0dba84a6cebc1528152ffc75f3da64
|
| 3 |
+
size 3320
|
deepseek-coder/shrunk-config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "/3fs-jd/prod/deepseek/shared/zhuqihao/public_model/deepseek-coder-7b-instruct2",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"hidden_act": "silu",
|
| 10 |
+
"hidden_size": 16,
|
| 11 |
+
"initializer_range": 0.02,
|
| 12 |
+
"intermediate_size": 32,
|
| 13 |
+
"max_position_embeddings": 128,
|
| 14 |
+
"model_type": "llama",
|
| 15 |
+
"num_attention_heads": 2,
|
| 16 |
+
"num_hidden_layers": 2,
|
| 17 |
+
"num_key_value_heads": 2,
|
| 18 |
+
"pretraining_tp": 1,
|
| 19 |
+
"rms_norm_eps": 1e-06,
|
| 20 |
+
"rope_scaling": {
|
| 21 |
+
"factor": 4.0,
|
| 22 |
+
"type": "linear"
|
| 23 |
+
},
|
| 24 |
+
"rope_theta": 100000,
|
| 25 |
+
"tie_word_embeddings": false,
|
| 26 |
+
"torch_dtype": "bfloat16",
|
| 27 |
+
"transformers_version": "4.34.1",
|
| 28 |
+
"use_cache": true,
|
| 29 |
+
"vocab_size": 128
|
| 30 |
+
}
|
deepseek-coder/source-config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "/3fs-jd/prod/deepseek/shared/zhuqihao/public_model/deepseek-coder-7b-instruct2",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"bos_token_id": 32013,
|
| 8 |
+
"eos_token_id": 32021,
|
| 9 |
+
"hidden_act": "silu",
|
| 10 |
+
"hidden_size": 4096,
|
| 11 |
+
"initializer_range": 0.02,
|
| 12 |
+
"intermediate_size": 11008,
|
| 13 |
+
"max_position_embeddings": 16384,
|
| 14 |
+
"model_type": "llama",
|
| 15 |
+
"num_attention_heads": 32,
|
| 16 |
+
"num_hidden_layers": 32,
|
| 17 |
+
"num_key_value_heads": 32,
|
| 18 |
+
"pretraining_tp": 1,
|
| 19 |
+
"rms_norm_eps": 1e-06,
|
| 20 |
+
"rope_scaling": {
|
| 21 |
+
"factor": 4.0,
|
| 22 |
+
"type": "linear"
|
| 23 |
+
},
|
| 24 |
+
"rope_theta": 100000,
|
| 25 |
+
"tie_word_embeddings": false,
|
| 26 |
+
"torch_dtype": "bfloat16",
|
| 27 |
+
"transformers_version": "4.34.1",
|
| 28 |
+
"use_cache": true,
|
| 29 |
+
"vocab_size": 32256
|
| 30 |
+
}
|
deepseek-coder/tokenizer/metadata.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 0,
|
| 3 |
+
"conceptual_token_id_range": [
|
| 4 |
+
0,
|
| 5 |
+
127
|
| 6 |
+
],
|
| 7 |
+
"disclaimer": "Synthetic metadata; it does not reproduce the upstream model tokenizer.",
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"eos_token_ids": [
|
| 10 |
+
1
|
| 11 |
+
],
|
| 12 |
+
"expected_eog_token_ids": [
|
| 13 |
+
1
|
| 14 |
+
],
|
| 15 |
+
"generator": "tools/add_gguf_q4km.py",
|
| 16 |
+
"pad_token_id": null,
|
| 17 |
+
"role": "synthetic auxiliary tokenizer for package conversion and smoke tests",
|
| 18 |
+
"schema_version": 1,
|
| 19 |
+
"sentencepiece_version": "0.2.2",
|
| 20 |
+
"source_tokenizer_reused": false,
|
| 21 |
+
"tokenizer_model_sha256": "0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee",
|
| 22 |
+
"unk_token_id": 0,
|
| 23 |
+
"vocab_size": 128
|
| 24 |
+
}
|
deepseek-coder/tokenizer/special_tokens_map.json
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token": "<unk>",
|
| 3 |
+
"eos_token": "<s>",
|
| 4 |
+
"unk_token": "<unk>"
|
| 5 |
+
}
|
deepseek-coder/tokenizer/tokenizer.model
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee
|
| 3 |
+
size 1980
|
deepseek-coder/tokenizer/tokenizer_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_bos_token": true,
|
| 3 |
+
"add_eos_token": false,
|
| 4 |
+
"add_prefix_space": true,
|
| 5 |
+
"bos_token": "<unk>",
|
| 6 |
+
"clean_up_tokenization_spaces": false,
|
| 7 |
+
"eos_token": "<s>",
|
| 8 |
+
"legacy": true,
|
| 9 |
+
"model_max_length": 128,
|
| 10 |
+
"tokenizer_class": "LlamaTokenizer",
|
| 11 |
+
"unk_token": "<unk>"
|
| 12 |
+
}
|
deepseek-coder/tokenizer/vocabulary.json
ADDED
|
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"0": 39,
|
| 3 |
+
"1": 40,
|
| 4 |
+
"2": 48,
|
| 5 |
+
"3": 41,
|
| 6 |
+
"4": 42,
|
| 7 |
+
"5": 43,
|
| 8 |
+
"6": 44,
|
| 9 |
+
"7": 45,
|
| 10 |
+
"8": 46,
|
| 11 |
+
"9": 47,
|
| 12 |
+
"<aux00>": 3,
|
| 13 |
+
"<aux01>": 4,
|
| 14 |
+
"<aux02>": 5,
|
| 15 |
+
"<aux03>": 6,
|
| 16 |
+
"<aux04>": 7,
|
| 17 |
+
"<aux05>": 8,
|
| 18 |
+
"<aux06>": 9,
|
| 19 |
+
"<aux07>": 10,
|
| 20 |
+
"<aux08>": 11,
|
| 21 |
+
"<aux09>": 12,
|
| 22 |
+
"<reserved02>": 2,
|
| 23 |
+
"<s>": 1,
|
| 24 |
+
"<unk>": 0,
|
| 25 |
+
"a": 124,
|
| 26 |
+
"b": 109,
|
| 27 |
+
"c": 118,
|
| 28 |
+
"d": 115,
|
| 29 |
+
"e": 125,
|
| 30 |
+
"f": 110,
|
| 31 |
+
"g": 108,
|
| 32 |
+
"h": 113,
|
| 33 |
+
"i": 117,
|
| 34 |
+
"k": 106,
|
| 35 |
+
"l": 122,
|
| 36 |
+
"m": 116,
|
| 37 |
+
"n": 120,
|
| 38 |
+
"o": 119,
|
| 39 |
+
"p": 112,
|
| 40 |
+
"q": 105,
|
| 41 |
+
"r": 114,
|
| 42 |
+
"s": 121,
|
| 43 |
+
"t": 123,
|
| 44 |
+
"u": 111,
|
| 45 |
+
"v": 103,
|
| 46 |
+
"w": 102,
|
| 47 |
+
"x": 101,
|
| 48 |
+
"y": 107,
|
| 49 |
+
"z": 104,
|
| 50 |
+
"\u2581": 126,
|
| 51 |
+
"\u258100": 49,
|
| 52 |
+
"\u258101": 99,
|
| 53 |
+
"\u2581012": 127,
|
| 54 |
+
"\u25810123456789": 13,
|
| 55 |
+
"\u258102": 50,
|
| 56 |
+
"\u258103": 51,
|
| 57 |
+
"\u258104": 52,
|
| 58 |
+
"\u258105": 53,
|
| 59 |
+
"\u258106": 54,
|
| 60 |
+
"\u258107": 55,
|
| 61 |
+
"\u258108": 56,
|
| 62 |
+
"\u258109": 57,
|
| 63 |
+
"\u258110": 58,
|
| 64 |
+
"\u258111": 59,
|
| 65 |
+
"\u258112": 60,
|
| 66 |
+
"\u258113": 61,
|
| 67 |
+
"\u258114": 62,
|
| 68 |
+
"\u258115": 63,
|
| 69 |
+
"\u258116": 64,
|
| 70 |
+
"\u258117": 65,
|
| 71 |
+
"\u258118": 66,
|
| 72 |
+
"\u258119": 67,
|
| 73 |
+
"\u258120": 68,
|
| 74 |
+
"\u258121": 69,
|
| 75 |
+
"\u258122": 70,
|
| 76 |
+
"\u258123": 71,
|
| 77 |
+
"\u258124": 72,
|
| 78 |
+
"\u258125": 73,
|
| 79 |
+
"\u258126": 74,
|
| 80 |
+
"\u258127": 75,
|
| 81 |
+
"\u258128": 76,
|
| 82 |
+
"\u258129": 77,
|
| 83 |
+
"\u258130": 78,
|
| 84 |
+
"\u258131": 79,
|
| 85 |
+
"\u258132": 80,
|
| 86 |
+
"\u258133": 81,
|
| 87 |
+
"\u258134": 82,
|
| 88 |
+
"\u258135": 83,
|
| 89 |
+
"\u258136": 84,
|
| 90 |
+
"\u258137": 85,
|
| 91 |
+
"\u258138": 86,
|
| 92 |
+
"\u258139": 87,
|
| 93 |
+
"\u258140": 88,
|
| 94 |
+
"\u258141": 89,
|
| 95 |
+
"\u258142": 90,
|
| 96 |
+
"\u258143": 91,
|
| 97 |
+
"\u258144": 92,
|
| 98 |
+
"\u258145": 93,
|
| 99 |
+
"\u258146": 94,
|
| 100 |
+
"\u258147": 95,
|
| 101 |
+
"\u258148": 96,
|
| 102 |
+
"\u258149": 97,
|
| 103 |
+
"\u258150": 98,
|
| 104 |
+
"\u258151": 100,
|
| 105 |
+
"\u2581alpha": 29,
|
| 106 |
+
"\u2581attention": 19,
|
| 107 |
+
"\u2581beta": 15,
|
| 108 |
+
"\u2581cache": 30,
|
| 109 |
+
"\u2581case": 14,
|
| 110 |
+
"\u2581context": 20,
|
| 111 |
+
"\u2581decode": 21,
|
| 112 |
+
"\u2581delta": 31,
|
| 113 |
+
"\u2581epsilon": 32,
|
| 114 |
+
"\u2581eta": 33,
|
| 115 |
+
"\u2581finite": 16,
|
| 116 |
+
"\u2581gamma": 34,
|
| 117 |
+
"\u2581key": 37,
|
| 118 |
+
"\u2581model": 22,
|
| 119 |
+
"\u2581prefill": 35,
|
| 120 |
+
"\u2581query": 17,
|
| 121 |
+
"\u2581small": 23,
|
| 122 |
+
"\u2581stable": 38,
|
| 123 |
+
"\u2581stream": 36,
|
| 124 |
+
"\u2581tensor": 24,
|
| 125 |
+
"\u2581theta": 25,
|
| 126 |
+
"\u2581token": 26,
|
| 127 |
+
"\u2581value": 27,
|
| 128 |
+
"\u2581weight": 28,
|
| 129 |
+
"\u2581zeta": 18
|
| 130 |
+
}
|
deepseek-coder/validation.json
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"repository": "deepseek-ai/deepseek-coder-6.7b-instruct",
|
| 3 |
+
"revision": "e5d64addd26a6a1db0f9b863abf6ee3141936807",
|
| 4 |
+
"source_config_sha256": "a3a2ce1a623b367d3799fb0c1e1ba12d5b8fa51a8e7abba14d10dd3739cd73e4",
|
| 5 |
+
"source_geometry": {
|
| 6 |
+
"hidden_size": 4096,
|
| 7 |
+
"intermediate_size": 11008,
|
| 8 |
+
"num_hidden_layers": 32,
|
| 9 |
+
"num_attention_heads": 32,
|
| 10 |
+
"num_key_value_heads": 32,
|
| 11 |
+
"vocab_size": 32256,
|
| 12 |
+
"max_position_embeddings": 16384
|
| 13 |
+
},
|
| 14 |
+
"shrunk_geometry": {
|
| 15 |
+
"hidden_size": 16,
|
| 16 |
+
"intermediate_size": 32,
|
| 17 |
+
"num_hidden_layers": 2,
|
| 18 |
+
"num_attention_heads": 2,
|
| 19 |
+
"num_key_value_heads": 2,
|
| 20 |
+
"vocab_size": 128,
|
| 21 |
+
"max_position_embeddings": 128
|
| 22 |
+
},
|
| 23 |
+
"modified_fields": [
|
| 24 |
+
"bos_token_id",
|
| 25 |
+
"eos_token_id",
|
| 26 |
+
"hidden_size",
|
| 27 |
+
"intermediate_size",
|
| 28 |
+
"max_position_embeddings",
|
| 29 |
+
"num_attention_heads",
|
| 30 |
+
"num_hidden_layers",
|
| 31 |
+
"num_key_value_heads",
|
| 32 |
+
"vocab_size"
|
| 33 |
+
],
|
| 34 |
+
"rules_triggered": [
|
| 35 |
+
"mha-preserved",
|
| 36 |
+
"out-of-range-special-tokens-remapped",
|
| 37 |
+
"rope-scaling-preserved"
|
| 38 |
+
],
|
| 39 |
+
"special_token_remap": {
|
| 40 |
+
"32013": 0,
|
| 41 |
+
"32021": 1
|
| 42 |
+
},
|
| 43 |
+
"generated_config_sha256": "1385d0151ad8bfe96845e0301e85e01cdab09ba16c11348e838754a2fb9505c9",
|
| 44 |
+
"validation_observations": [],
|
| 45 |
+
"generated_parameter_count": 9296,
|
| 46 |
+
"generated_checkpoint_sha256": "cba3bb2298077f3bdeee9da78a4d0ae7586c3a3964e7a40eb2f9e5dc2631c4cc",
|
| 47 |
+
"generated_checkpoint_size": 39288,
|
| 48 |
+
"reload_comparison": {
|
| 49 |
+
"prefill_logits": {
|
| 50 |
+
"max_absolute_error": 0.0,
|
| 51 |
+
"max_relative_error": 0.0,
|
| 52 |
+
"rtol": 1e-06,
|
| 53 |
+
"atol": 1e-07
|
| 54 |
+
},
|
| 55 |
+
"decode_logits": {
|
| 56 |
+
"max_absolute_error": 1.4901161193847656e-08,
|
| 57 |
+
"max_relative_error": 1.2509694897744339e-05,
|
| 58 |
+
"rtol": 1e-06,
|
| 59 |
+
"atol": 1e-07
|
| 60 |
+
},
|
| 61 |
+
"cache_free_decode_logits": {
|
| 62 |
+
"max_absolute_error": 0.0,
|
| 63 |
+
"max_relative_error": 0.0,
|
| 64 |
+
"rtol": 1e-06,
|
| 65 |
+
"atol": 1e-07
|
| 66 |
+
}
|
| 67 |
+
},
|
| 68 |
+
"artifact_size": 46660,
|
| 69 |
+
"state": "POSTCHECK_OK",
|
| 70 |
+
"failure_stage": null,
|
| 71 |
+
"exception": null
|
| 72 |
+
}
|
hermes3-llama31/case.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architecture": "llama",
|
| 3 |
+
"artifact_id": "hermes3-llama31",
|
| 4 |
+
"auxiliary_tokenizer": {
|
| 5 |
+
"metadata": "tokenizer/metadata.json",
|
| 6 |
+
"path": "tokenizer"
|
| 7 |
+
},
|
| 8 |
+
"comparison_policy": {
|
| 9 |
+
"cached_vs_cache_free": {
|
| 10 |
+
"atol": 1e-06,
|
| 11 |
+
"rtol": 1e-05
|
| 12 |
+
},
|
| 13 |
+
"primary_value": "logits",
|
| 14 |
+
"reload": {
|
| 15 |
+
"atol": 1e-07,
|
| 16 |
+
"rtol": 1e-06
|
| 17 |
+
},
|
| 18 |
+
"require_finite": true
|
| 19 |
+
},
|
| 20 |
+
"format": {
|
| 21 |
+
"container": "huggingface-safetensors",
|
| 22 |
+
"dtype": "float32",
|
| 23 |
+
"layout": "single-file",
|
| 24 |
+
"package_path": "package"
|
| 25 |
+
},
|
| 26 |
+
"packages": {
|
| 27 |
+
"gguf_q4_k_m": {
|
| 28 |
+
"metadata": "gguf/metadata.json",
|
| 29 |
+
"model": "gguf/model-Q4_K_M.gguf",
|
| 30 |
+
"reference": "gguf/reference.json"
|
| 31 |
+
},
|
| 32 |
+
"safetensors_f32": {
|
| 33 |
+
"config": "package/config.json",
|
| 34 |
+
"model": "package/model.safetensors",
|
| 35 |
+
"sha256": "22d5f7094e33c1e50eb689b20f56094239f0df30a23b17ef508ce1a6b49c81dc",
|
| 36 |
+
"size_bytes": 347488
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"reference_runtime": {
|
| 40 |
+
"config_class": "LlamaConfig",
|
| 41 |
+
"implementation": "huggingface-transformers",
|
| 42 |
+
"model_class": "LlamaForCausalLM",
|
| 43 |
+
"trust_remote_code": false,
|
| 44 |
+
"version": "4.55.0"
|
| 45 |
+
},
|
| 46 |
+
"schema_version": 1,
|
| 47 |
+
"semantic_features": {
|
| 48 |
+
"explicit_head_dim": false,
|
| 49 |
+
"non_null_rope_scaling": true,
|
| 50 |
+
"query_width_matches_hidden_size": true
|
| 51 |
+
},
|
| 52 |
+
"source": {
|
| 53 |
+
"config_sha256": "7e189a0da074da6a878d5a276336caf3d62c57138f461bc6239402639f1bc6f5",
|
| 54 |
+
"repository": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 55 |
+
"revision": "896ea440e5a9e6070e3d8a2774daf2b481ab425b"
|
| 56 |
+
},
|
| 57 |
+
"validation": "POSTCHECK_OK"
|
| 58 |
+
}
|
hermes3-llama31/config-diff.json
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"path": "/bos_token_id",
|
| 4 |
+
"source": 128000,
|
| 5 |
+
"shrunk": 0
|
| 6 |
+
},
|
| 7 |
+
{
|
| 8 |
+
"path": "/eos_token_id",
|
| 9 |
+
"source": 128040,
|
| 10 |
+
"shrunk": 1
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"path": "/hidden_size",
|
| 14 |
+
"source": 4096,
|
| 15 |
+
"shrunk": 64
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"path": "/intermediate_size",
|
| 19 |
+
"source": 14336,
|
| 20 |
+
"shrunk": 128
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"path": "/max_position_embeddings",
|
| 24 |
+
"source": 131072,
|
| 25 |
+
"shrunk": 128
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"path": "/num_attention_heads",
|
| 29 |
+
"source": 32,
|
| 30 |
+
"shrunk": 8
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"path": "/num_hidden_layers",
|
| 34 |
+
"source": 32,
|
| 35 |
+
"shrunk": 2
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"path": "/num_key_value_heads",
|
| 39 |
+
"source": 8,
|
| 40 |
+
"shrunk": 2
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"path": "/vocab_size",
|
| 44 |
+
"source": 128256,
|
| 45 |
+
"shrunk": 128
|
| 46 |
+
}
|
| 47 |
+
]
|
hermes3-llama31/gguf/metadata.json
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"commands": {
|
| 3 |
+
"convert": [
|
| 4 |
+
"PYTHON",
|
| 5 |
+
"LLAMA_CPP/convert_hf_to_gguf.py",
|
| 6 |
+
"STAGING",
|
| 7 |
+
"--outfile",
|
| 8 |
+
"INTERMEDIATE",
|
| 9 |
+
"--outtype",
|
| 10 |
+
"f16"
|
| 11 |
+
],
|
| 12 |
+
"quantize": [
|
| 13 |
+
"LLAMA_QUANTIZE",
|
| 14 |
+
"--output-tensor-type",
|
| 15 |
+
"f16",
|
| 16 |
+
"--token-embedding-type",
|
| 17 |
+
"f16",
|
| 18 |
+
"INTERMEDIATE",
|
| 19 |
+
"OUTPUT",
|
| 20 |
+
"Q4_K_M"
|
| 21 |
+
]
|
| 22 |
+
},
|
| 23 |
+
"comparison_to_f32_reference": {
|
| 24 |
+
"decode": {
|
| 25 |
+
"max_abs_error": 0.0047625526785850525,
|
| 26 |
+
"mean_abs_error": 0.0015001886058598757,
|
| 27 |
+
"top1_agreement": true,
|
| 28 |
+
"top5_overlap": 5
|
| 29 |
+
},
|
| 30 |
+
"prefill": {
|
| 31 |
+
"max_abs_error": 0.006160363554954529,
|
| 32 |
+
"mean_abs_error": 0.0017008509021252394,
|
| 33 |
+
"top1_agreement": true,
|
| 34 |
+
"top5_overlap": 5
|
| 35 |
+
}
|
| 36 |
+
},
|
| 37 |
+
"converter": "convert_hf_to_gguf.py",
|
| 38 |
+
"converter_log_sha256": "88aec1059a6ef4c697bd76d1e40c6f33ebb06801dd6deef3ed5145c4b52c1de4",
|
| 39 |
+
"direct_token_validation": {
|
| 40 |
+
"all_logits_finite": true,
|
| 41 |
+
"decode_token_id": 13,
|
| 42 |
+
"exactly_reproducible": true,
|
| 43 |
+
"prefill_token_ids": [
|
| 44 |
+
3,
|
| 45 |
+
5,
|
| 46 |
+
7,
|
| 47 |
+
11
|
| 48 |
+
]
|
| 49 |
+
},
|
| 50 |
+
"effective_eog_token_ids": [
|
| 51 |
+
1
|
| 52 |
+
],
|
| 53 |
+
"eog_validation_range": [
|
| 54 |
+
0,
|
| 55 |
+
127
|
| 56 |
+
],
|
| 57 |
+
"format": "GGUF",
|
| 58 |
+
"intermediate_f16_sha256": "142b77f9d5bc8393c5e15e049e51de92562ad5e91583c8e985734eb15571e346",
|
| 59 |
+
"intermediate_retained": false,
|
| 60 |
+
"llama_cpp_revision": "40b740ad05c531b9d57aca6698c3ed553a9e784c",
|
| 61 |
+
"model_sha256": "6b53e66bd258178cb1ceb6f89ad36a3b909e912c430ec5ccac06a476d6a652e4",
|
| 62 |
+
"model_size_bytes": 90688,
|
| 63 |
+
"notes": "Q4_K_M is lossy; numerical differences are recorded as observations, not pass thresholds.",
|
| 64 |
+
"quantization_requested": "Q4_K_M",
|
| 65 |
+
"quantizer": "llama-quantize",
|
| 66 |
+
"quantizer_fallback_observed": true,
|
| 67 |
+
"quantizer_log_sha256": "21f026628b11158d47bbbbfff2eca907d8ac19f6780291b70cc614eb4802f751",
|
| 68 |
+
"schema_version": 1,
|
| 69 |
+
"source_auxiliary_tokenizer_sha256": "0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee",
|
| 70 |
+
"source_package_model_sha256": "22d5f7094e33c1e50eb689b20f56094239f0df30a23b17ef508ce1a6b49c81dc",
|
| 71 |
+
"staging_special_token_normalization": {
|
| 72 |
+
"applied": false,
|
| 73 |
+
"gguf_primary_eos_token_id": 1,
|
| 74 |
+
"reason": "GGUF exposes one primary EOS metadata field; all source EOS IDs remain documented by tokenizer metadata.",
|
| 75 |
+
"source_eos_token_id": 1
|
| 76 |
+
},
|
| 77 |
+
"tensor_type_histogram": {
|
| 78 |
+
"F16": 2,
|
| 79 |
+
"F32": 6,
|
| 80 |
+
"Q5_0": 12,
|
| 81 |
+
"Q8_0": 2
|
| 82 |
+
},
|
| 83 |
+
"tiny_dimension_tensor_overrides": {
|
| 84 |
+
"output.weight": "F16",
|
| 85 |
+
"reason": "These tensors can be too narrow for the Q8_0 types selected by the Q4_K_M mixture.",
|
| 86 |
+
"token_embd.weight": "F16"
|
| 87 |
+
},
|
| 88 |
+
"tokenizer_smoke": {
|
| 89 |
+
"hf_native_ids_equal": true,
|
| 90 |
+
"text": "a",
|
| 91 |
+
"token_ids": [
|
| 92 |
+
126,
|
| 93 |
+
124
|
| 94 |
+
]
|
| 95 |
+
}
|
| 96 |
+
}
|
hermes3-llama31/gguf/model-Q4_K_M.gguf
ADDED
|
Binary file (90.7 kB). View file
|
|
|
hermes3-llama31/gguf/reference.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"model_type": "llama",
|
| 4 |
+
"file_type": 15,
|
| 5 |
+
"tokenizer_model": "llama",
|
| 6 |
+
"vocab_size": 128,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"pad_token_id": -1,
|
| 10 |
+
"declared_bos_token_id": 0,
|
| 11 |
+
"declared_eos_token_id": 1,
|
| 12 |
+
"declared_pad_token_id": -1,
|
| 13 |
+
"eog_token_ids": [1],
|
| 14 |
+
"prefill_token_ids": [3,5,7,11],
|
| 15 |
+
"decode_token_id": 13,
|
| 16 |
+
"smoke_text": "a",
|
| 17 |
+
"smoke_token_ids": [126,124],
|
| 18 |
+
"all_logits_finite": true,
|
| 19 |
+
"prefill_last_logits": [-0.00373650342,0.0142230652,0.0239198729,-0.0922453254,-0.00640527904,0.106421955,0.0303641409,0.134387836,-0.0646145046,-0.0326549113,-0.266988128,0.0424443856,0.261955857,0.0742194802,0.020770099,-0.0110064894,-0.0566034429,0.081981875,0.0100149941,0.150510162,0.000316821039,-0.0865940452,0.191807479,0.12882933,0.0606084391,-0.118425027,0.133052945,0.26927352,0.202962831,-0.174355865,0.0954167545,-0.0601493679,-0.0892238021,0.290301144,-0.170065135,-0.0126270521,-0.0341819823,-0.105074942,-0.118038021,0.0656491369,0.168856815,0.102200747,-0.0431903526,-0.0291026086,0.00605118275,-0.137299255,-0.0447447039,0.0610228106,0.11029774,0.0509579405,-0.113215968,-0.111768439,0.140373424,0.0857898295,-0.0617344901,-0.0983991623,-0.0256835967,0.141078994,-0.0503700189,-0.0851275399,0.0288536511,0.0210943595,-0.019230729,-0.181046665,-0.0119840316,-0.0243329406,-0.0487618782,0.0532716662,-0.191456199,0.000374857336,-0.0456871837,0.141764849,0.0335974731,0.0139211975,0.132296681,0.0142042711,-0.0316513069,0.0576448403,-0.0662858784,0.0404601246,0.106457949,-0.0576132387,0.0832884312,0.0936105251,0.0264142081,-0.0835624933,-0.0219739843,0.0318763927,0.045923844,-0.18683444,-0.0602085665,0.115002647,-0.151682347,0.0406073518,-0.066142723,-0.000689655542,-0.00596629083,0.220412403,0.0790107474,-0.0410532169,0.106174447,0.220154881,0.0211779699,-0.00975717232,0.0585278198,0.101865157,-0.139449969,-0.0043367222,0.117498249,-0.0502359346,-0.0348631367,0.0061924085,0.0383126363,-0.0773586631,-0.0417589732,0.239479259,-0.0797864273,0.102656394,0.135014996,0.0186934173,0.105619512,-0.0417982079,-0.0136995539,-0.21335578,-0.205111086,0.183098704,0.009346582,-0.161455452],
|
| 20 |
+
"decode_logits": [-0.00337634049,0.0726294369,-0.0141554587,-0.247402698,0.0417947732,0.195167959,0.190877706,0.040277306,0.0169912614,-0.0830896646,-0.233932301,-0.140349105,0.15574877,0.0245330632,0.0259653404,0.0658680722,0.0165881291,-0.072418347,-0.15179655,-0.00675424188,-0.00137612224,0.0735251978,0.100361645,0.117904626,0.0716115236,-0.0685327947,-0.0370465443,0.079740718,-0.0572876856,-0.0507836118,0.169267833,-0.0983380079,-0.148488611,0.180731505,0.0411532857,0.00140581839,-0.0213001631,-0.0688719079,-0.230596095,0.0912246853,0.107642263,0.213849753,0.02156749,0.0259810649,-0.116949573,-0.0996030495,0.0043765828,-0.0261323266,-0.0125979707,-0.0462644286,-0.116449215,-0.11063502,0.104174137,-0.137514234,-0.0188461244,-0.0477778651,-0.113706917,0.0612858497,-0.197109967,-0.00296059251,0.0926772505,-0.101441629,-0.105332702,-0.091951862,-0.111329332,-0.0282375067,-0.0399259366,-0.121300526,-0.226675123,-0.2248106,0.00103991479,0.0117659364,0.108930476,-0.00467790104,0.0216557905,-0.120729253,-0.00483879074,0.174247563,-0.0620732568,0.0687803775,-0.100114822,-0.0408749729,0.0993463546,0.0793903172,0.0867445692,0.0702011213,0.0179631077,-0.00652953237,-0.0805617049,-0.0984117687,-0.131118998,-0.111845583,-0.115344718,0.124632411,0.0136053152,-0.0263243206,0.117817819,0.059105929,0.177215949,-0.116577491,-0.0306722298,0.153529868,0.119137377,-0.140856877,-0.048977226,0.0478735715,0.0681472868,0.020005133,-0.048132658,-0.135485604,0.126471475,0.121675432,-0.142729908,0.00536482781,-0.0212973878,0.0230659768,-0.0304893833,0.0886625201,0.15632385,0.131743997,0.116146527,0.0367615558,0.00405740738,0.0132997856,-0.151548445,0.00792044029,0.0373035111,0.0708587989]
|
| 21 |
+
}
|
hermes3-llama31/inputs.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b49b7cca71e662452eb97b184578af4f59e9a93bc541d621986836cd7e2c395
|
| 3 |
+
size 176
|
hermes3-llama31/package/config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 0,
|
| 9 |
+
"eos_token_id": 1,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 64,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 128,
|
| 14 |
+
"max_position_embeddings": 128,
|
| 15 |
+
"mlp_bias": false,
|
| 16 |
+
"model_type": "llama",
|
| 17 |
+
"num_attention_heads": 8,
|
| 18 |
+
"num_hidden_layers": 2,
|
| 19 |
+
"num_key_value_heads": 2,
|
| 20 |
+
"pretraining_tp": 1,
|
| 21 |
+
"rms_norm_eps": 1e-05,
|
| 22 |
+
"rope_scaling": {
|
| 23 |
+
"factor": 8.0,
|
| 24 |
+
"high_freq_factor": 4.0,
|
| 25 |
+
"low_freq_factor": 1.0,
|
| 26 |
+
"original_max_position_embeddings": 8192,
|
| 27 |
+
"rope_type": "llama3"
|
| 28 |
+
},
|
| 29 |
+
"rope_theta": 500000.0,
|
| 30 |
+
"tie_word_embeddings": false,
|
| 31 |
+
"torch_dtype": "bfloat16",
|
| 32 |
+
"transformers_version": "4.44.0.dev0",
|
| 33 |
+
"use_cache": true,
|
| 34 |
+
"vocab_size": 128
|
| 35 |
+
}
|
hermes3-llama31/package/generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 0,
|
| 4 |
+
"eos_token_id": 1,
|
| 5 |
+
"transformers_version": "4.55.0"
|
| 6 |
+
}
|
hermes3-llama31/package/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:22d5f7094e33c1e50eb689b20f56094239f0df30a23b17ef508ce1a6b49c81dc
|
| 3 |
+
size 347488
|
hermes3-llama31/provenance.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"repository": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 4 |
+
"revision": "896ea440e5a9e6070e3d8a2774daf2b481ab425b",
|
| 5 |
+
"source_config_sha256": "7e189a0da074da6a878d5a276336caf3d62c57138f461bc6239402639f1bc6f5",
|
| 6 |
+
"shrinker_version": "llama-preservation-v1",
|
| 7 |
+
"conf_tool_version": "0.2.0",
|
| 8 |
+
"seed": 4850181387854826561,
|
| 9 |
+
"rng": "tlfloat::LCG64 from testerutil.hpp",
|
| 10 |
+
"tlfloat_revision": "936f0b093f0a79c0534fe5185ff80b6a649c576c",
|
| 11 |
+
"generated_config_sha256": "5ddca639398c8e5571a4d8670ea1b111a78c1cdb4342eafb22c68abb1325efcc",
|
| 12 |
+
"generated_checkpoint_sha256": "22d5f7094e33c1e50eb689b20f56094239f0df30a23b17ef508ce1a6b49c81dc",
|
| 13 |
+
"python": "3.12.13",
|
| 14 |
+
"torch": "2.8.0+cu128",
|
| 15 |
+
"transformers": "4.55.0",
|
| 16 |
+
"accelerate": "1.10.0",
|
| 17 |
+
"safetensors": "0.6.2",
|
| 18 |
+
"huggingface_hub": "0.34.4"
|
| 19 |
+
}
|
hermes3-llama31/reference/outputs.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:97918f673a3659e9cff825ec69b974cb6d8ef2cd8e19da0c4f3da2db0363f69d
|
| 3 |
+
size 3320
|
hermes3-llama31/shrunk-config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 0,
|
| 9 |
+
"eos_token_id": 1,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 64,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 128,
|
| 14 |
+
"max_position_embeddings": 128,
|
| 15 |
+
"mlp_bias": false,
|
| 16 |
+
"model_type": "llama",
|
| 17 |
+
"num_attention_heads": 8,
|
| 18 |
+
"num_hidden_layers": 2,
|
| 19 |
+
"num_key_value_heads": 2,
|
| 20 |
+
"pretraining_tp": 1,
|
| 21 |
+
"rms_norm_eps": 1e-05,
|
| 22 |
+
"rope_scaling": {
|
| 23 |
+
"factor": 8.0,
|
| 24 |
+
"high_freq_factor": 4.0,
|
| 25 |
+
"low_freq_factor": 1.0,
|
| 26 |
+
"original_max_position_embeddings": 8192,
|
| 27 |
+
"rope_type": "llama3"
|
| 28 |
+
},
|
| 29 |
+
"rope_theta": 500000.0,
|
| 30 |
+
"tie_word_embeddings": false,
|
| 31 |
+
"torch_dtype": "bfloat16",
|
| 32 |
+
"transformers_version": "4.44.0.dev0",
|
| 33 |
+
"use_cache": true,
|
| 34 |
+
"vocab_size": 128
|
| 35 |
+
}
|
hermes3-llama31/source-config.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 128000,
|
| 9 |
+
"eos_token_id": 128040,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 4096,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 14336,
|
| 14 |
+
"max_position_embeddings": 131072,
|
| 15 |
+
"mlp_bias": false,
|
| 16 |
+
"model_type": "llama",
|
| 17 |
+
"num_attention_heads": 32,
|
| 18 |
+
"num_hidden_layers": 32,
|
| 19 |
+
"num_key_value_heads": 8,
|
| 20 |
+
"pretraining_tp": 1,
|
| 21 |
+
"rms_norm_eps": 1e-05,
|
| 22 |
+
"rope_scaling": {
|
| 23 |
+
"factor": 8.0,
|
| 24 |
+
"high_freq_factor": 4.0,
|
| 25 |
+
"low_freq_factor": 1.0,
|
| 26 |
+
"original_max_position_embeddings": 8192,
|
| 27 |
+
"rope_type": "llama3"
|
| 28 |
+
},
|
| 29 |
+
"rope_theta": 500000.0,
|
| 30 |
+
"tie_word_embeddings": false,
|
| 31 |
+
"torch_dtype": "bfloat16",
|
| 32 |
+
"transformers_version": "4.44.0.dev0",
|
| 33 |
+
"use_cache": true,
|
| 34 |
+
"vocab_size": 128256
|
| 35 |
+
}
|
hermes3-llama31/tokenizer/metadata.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 0,
|
| 3 |
+
"conceptual_token_id_range": [
|
| 4 |
+
0,
|
| 5 |
+
127
|
| 6 |
+
],
|
| 7 |
+
"disclaimer": "Synthetic metadata; it does not reproduce the upstream model tokenizer.",
|
| 8 |
+
"eos_token_id": 1,
|
| 9 |
+
"eos_token_ids": [
|
| 10 |
+
1
|
| 11 |
+
],
|
| 12 |
+
"expected_eog_token_ids": [
|
| 13 |
+
1
|
| 14 |
+
],
|
| 15 |
+
"generator": "tools/add_gguf_q4km.py",
|
| 16 |
+
"pad_token_id": null,
|
| 17 |
+
"role": "synthetic auxiliary tokenizer for package conversion and smoke tests",
|
| 18 |
+
"schema_version": 1,
|
| 19 |
+
"sentencepiece_version": "0.2.2",
|
| 20 |
+
"source_tokenizer_reused": false,
|
| 21 |
+
"tokenizer_model_sha256": "0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee",
|
| 22 |
+
"unk_token_id": 0,
|
| 23 |
+
"vocab_size": 128
|
| 24 |
+
}
|
hermes3-llama31/tokenizer/special_tokens_map.json
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token": "<unk>",
|
| 3 |
+
"eos_token": "<s>",
|
| 4 |
+
"unk_token": "<unk>"
|
| 5 |
+
}
|
hermes3-llama31/tokenizer/tokenizer.model
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee
|
| 3 |
+
size 1980
|
hermes3-llama31/tokenizer/tokenizer_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_bos_token": true,
|
| 3 |
+
"add_eos_token": false,
|
| 4 |
+
"add_prefix_space": true,
|
| 5 |
+
"bos_token": "<unk>",
|
| 6 |
+
"clean_up_tokenization_spaces": false,
|
| 7 |
+
"eos_token": "<s>",
|
| 8 |
+
"legacy": true,
|
| 9 |
+
"model_max_length": 128,
|
| 10 |
+
"tokenizer_class": "LlamaTokenizer",
|
| 11 |
+
"unk_token": "<unk>"
|
| 12 |
+
}
|
hermes3-llama31/tokenizer/vocabulary.json
ADDED
|
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"0": 39,
|
| 3 |
+
"1": 40,
|
| 4 |
+
"2": 48,
|
| 5 |
+
"3": 41,
|
| 6 |
+
"4": 42,
|
| 7 |
+
"5": 43,
|
| 8 |
+
"6": 44,
|
| 9 |
+
"7": 45,
|
| 10 |
+
"8": 46,
|
| 11 |
+
"9": 47,
|
| 12 |
+
"<aux00>": 3,
|
| 13 |
+
"<aux01>": 4,
|
| 14 |
+
"<aux02>": 5,
|
| 15 |
+
"<aux03>": 6,
|
| 16 |
+
"<aux04>": 7,
|
| 17 |
+
"<aux05>": 8,
|
| 18 |
+
"<aux06>": 9,
|
| 19 |
+
"<aux07>": 10,
|
| 20 |
+
"<aux08>": 11,
|
| 21 |
+
"<aux09>": 12,
|
| 22 |
+
"<reserved02>": 2,
|
| 23 |
+
"<s>": 1,
|
| 24 |
+
"<unk>": 0,
|
| 25 |
+
"a": 124,
|
| 26 |
+
"b": 109,
|
| 27 |
+
"c": 118,
|
| 28 |
+
"d": 115,
|
| 29 |
+
"e": 125,
|
| 30 |
+
"f": 110,
|
| 31 |
+
"g": 108,
|
| 32 |
+
"h": 113,
|
| 33 |
+
"i": 117,
|
| 34 |
+
"k": 106,
|
| 35 |
+
"l": 122,
|
| 36 |
+
"m": 116,
|
| 37 |
+
"n": 120,
|
| 38 |
+
"o": 119,
|
| 39 |
+
"p": 112,
|
| 40 |
+
"q": 105,
|
| 41 |
+
"r": 114,
|
| 42 |
+
"s": 121,
|
| 43 |
+
"t": 123,
|
| 44 |
+
"u": 111,
|
| 45 |
+
"v": 103,
|
| 46 |
+
"w": 102,
|
| 47 |
+
"x": 101,
|
| 48 |
+
"y": 107,
|
| 49 |
+
"z": 104,
|
| 50 |
+
"\u2581": 126,
|
| 51 |
+
"\u258100": 49,
|
| 52 |
+
"\u258101": 99,
|
| 53 |
+
"\u2581012": 127,
|
| 54 |
+
"\u25810123456789": 13,
|
| 55 |
+
"\u258102": 50,
|
| 56 |
+
"\u258103": 51,
|
| 57 |
+
"\u258104": 52,
|
| 58 |
+
"\u258105": 53,
|
| 59 |
+
"\u258106": 54,
|
| 60 |
+
"\u258107": 55,
|
| 61 |
+
"\u258108": 56,
|
| 62 |
+
"\u258109": 57,
|
| 63 |
+
"\u258110": 58,
|
| 64 |
+
"\u258111": 59,
|
| 65 |
+
"\u258112": 60,
|
| 66 |
+
"\u258113": 61,
|
| 67 |
+
"\u258114": 62,
|
| 68 |
+
"\u258115": 63,
|
| 69 |
+
"\u258116": 64,
|
| 70 |
+
"\u258117": 65,
|
| 71 |
+
"\u258118": 66,
|
| 72 |
+
"\u258119": 67,
|
| 73 |
+
"\u258120": 68,
|
| 74 |
+
"\u258121": 69,
|
| 75 |
+
"\u258122": 70,
|
| 76 |
+
"\u258123": 71,
|
| 77 |
+
"\u258124": 72,
|
| 78 |
+
"\u258125": 73,
|
| 79 |
+
"\u258126": 74,
|
| 80 |
+
"\u258127": 75,
|
| 81 |
+
"\u258128": 76,
|
| 82 |
+
"\u258129": 77,
|
| 83 |
+
"\u258130": 78,
|
| 84 |
+
"\u258131": 79,
|
| 85 |
+
"\u258132": 80,
|
| 86 |
+
"\u258133": 81,
|
| 87 |
+
"\u258134": 82,
|
| 88 |
+
"\u258135": 83,
|
| 89 |
+
"\u258136": 84,
|
| 90 |
+
"\u258137": 85,
|
| 91 |
+
"\u258138": 86,
|
| 92 |
+
"\u258139": 87,
|
| 93 |
+
"\u258140": 88,
|
| 94 |
+
"\u258141": 89,
|
| 95 |
+
"\u258142": 90,
|
| 96 |
+
"\u258143": 91,
|
| 97 |
+
"\u258144": 92,
|
| 98 |
+
"\u258145": 93,
|
| 99 |
+
"\u258146": 94,
|
| 100 |
+
"\u258147": 95,
|
| 101 |
+
"\u258148": 96,
|
| 102 |
+
"\u258149": 97,
|
| 103 |
+
"\u258150": 98,
|
| 104 |
+
"\u258151": 100,
|
| 105 |
+
"\u2581alpha": 29,
|
| 106 |
+
"\u2581attention": 19,
|
| 107 |
+
"\u2581beta": 15,
|
| 108 |
+
"\u2581cache": 30,
|
| 109 |
+
"\u2581case": 14,
|
| 110 |
+
"\u2581context": 20,
|
| 111 |
+
"\u2581decode": 21,
|
| 112 |
+
"\u2581delta": 31,
|
| 113 |
+
"\u2581epsilon": 32,
|
| 114 |
+
"\u2581eta": 33,
|
| 115 |
+
"\u2581finite": 16,
|
| 116 |
+
"\u2581gamma": 34,
|
| 117 |
+
"\u2581key": 37,
|
| 118 |
+
"\u2581model": 22,
|
| 119 |
+
"\u2581prefill": 35,
|
| 120 |
+
"\u2581query": 17,
|
| 121 |
+
"\u2581small": 23,
|
| 122 |
+
"\u2581stable": 38,
|
| 123 |
+
"\u2581stream": 36,
|
| 124 |
+
"\u2581tensor": 24,
|
| 125 |
+
"\u2581theta": 25,
|
| 126 |
+
"\u2581token": 26,
|
| 127 |
+
"\u2581value": 27,
|
| 128 |
+
"\u2581weight": 28,
|
| 129 |
+
"\u2581zeta": 18
|
| 130 |
+
}
|
hermes3-llama31/validation.json
ADDED
|
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"repository": "NousResearch/Hermes-3-Llama-3.1-8B",
|
| 3 |
+
"revision": "896ea440e5a9e6070e3d8a2774daf2b481ab425b",
|
| 4 |
+
"source_config_sha256": "7e189a0da074da6a878d5a276336caf3d62c57138f461bc6239402639f1bc6f5",
|
| 5 |
+
"source_geometry": {
|
| 6 |
+
"hidden_size": 4096,
|
| 7 |
+
"intermediate_size": 14336,
|
| 8 |
+
"num_hidden_layers": 32,
|
| 9 |
+
"num_attention_heads": 32,
|
| 10 |
+
"num_key_value_heads": 8,
|
| 11 |
+
"vocab_size": 128256,
|
| 12 |
+
"max_position_embeddings": 131072
|
| 13 |
+
},
|
| 14 |
+
"shrunk_geometry": {
|
| 15 |
+
"hidden_size": 64,
|
| 16 |
+
"intermediate_size": 128,
|
| 17 |
+
"num_hidden_layers": 2,
|
| 18 |
+
"num_attention_heads": 8,
|
| 19 |
+
"num_key_value_heads": 2,
|
| 20 |
+
"vocab_size": 128,
|
| 21 |
+
"max_position_embeddings": 128
|
| 22 |
+
},
|
| 23 |
+
"modified_fields": [
|
| 24 |
+
"bos_token_id",
|
| 25 |
+
"eos_token_id",
|
| 26 |
+
"hidden_size",
|
| 27 |
+
"intermediate_size",
|
| 28 |
+
"max_position_embeddings",
|
| 29 |
+
"num_attention_heads",
|
| 30 |
+
"num_hidden_layers",
|
| 31 |
+
"num_key_value_heads",
|
| 32 |
+
"vocab_size"
|
| 33 |
+
],
|
| 34 |
+
"rules_triggered": [
|
| 35 |
+
"gqa-ratio-4-preserved",
|
| 36 |
+
"out-of-range-special-tokens-remapped",
|
| 37 |
+
"rope-scaling-preserved"
|
| 38 |
+
],
|
| 39 |
+
"special_token_remap": {
|
| 40 |
+
"128000": 0,
|
| 41 |
+
"128040": 1
|
| 42 |
+
},
|
| 43 |
+
"generated_config_sha256": "5ddca639398c8e5571a4d8670ea1b111a78c1cdb4342eafb22c68abb1325efcc",
|
| 44 |
+
"validation_observations": [
|
| 45 |
+
"Transformers warns that rope_scaling.original_max_position_embeddings is not less than max_position_embeddings; complete post-check still passes"
|
| 46 |
+
],
|
| 47 |
+
"generated_parameter_count": 86336,
|
| 48 |
+
"generated_checkpoint_sha256": "22d5f7094e33c1e50eb689b20f56094239f0df30a23b17ef508ce1a6b49c81dc",
|
| 49 |
+
"generated_checkpoint_size": 347488,
|
| 50 |
+
"reload_comparison": {
|
| 51 |
+
"prefill_logits": {
|
| 52 |
+
"max_absolute_error": 0.0,
|
| 53 |
+
"max_relative_error": 0.0,
|
| 54 |
+
"rtol": 1e-06,
|
| 55 |
+
"atol": 1e-07
|
| 56 |
+
},
|
| 57 |
+
"decode_logits": {
|
| 58 |
+
"max_absolute_error": 0.0,
|
| 59 |
+
"max_relative_error": 0.0,
|
| 60 |
+
"rtol": 1e-06,
|
| 61 |
+
"atol": 1e-07
|
| 62 |
+
},
|
| 63 |
+
"cache_free_decode_logits": {
|
| 64 |
+
"max_absolute_error": 0.0,
|
| 65 |
+
"max_relative_error": 0.0,
|
| 66 |
+
"rtol": 1e-06,
|
| 67 |
+
"atol": 1e-07
|
| 68 |
+
}
|
| 69 |
+
},
|
| 70 |
+
"artifact_size": 355223,
|
| 71 |
+
"state": "POSTCHECK_OK",
|
| 72 |
+
"failure_stage": null,
|
| 73 |
+
"exception": null
|
| 74 |
+
}
|
livekit-turn-detector/case.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architecture": "llama",
|
| 3 |
+
"artifact_id": "livekit-turn-detector",
|
| 4 |
+
"auxiliary_tokenizer": {
|
| 5 |
+
"metadata": "tokenizer/metadata.json",
|
| 6 |
+
"path": "tokenizer"
|
| 7 |
+
},
|
| 8 |
+
"comparison_policy": {
|
| 9 |
+
"cached_vs_cache_free": {
|
| 10 |
+
"atol": 1e-06,
|
| 11 |
+
"rtol": 1e-05
|
| 12 |
+
},
|
| 13 |
+
"primary_value": "logits",
|
| 14 |
+
"reload": {
|
| 15 |
+
"atol": 1e-07,
|
| 16 |
+
"rtol": 1e-06
|
| 17 |
+
},
|
| 18 |
+
"require_finite": true
|
| 19 |
+
},
|
| 20 |
+
"format": {
|
| 21 |
+
"container": "huggingface-safetensors",
|
| 22 |
+
"dtype": "float32",
|
| 23 |
+
"layout": "single-file",
|
| 24 |
+
"package_path": "package"
|
| 25 |
+
},
|
| 26 |
+
"packages": {
|
| 27 |
+
"gguf_q4_k_m": {
|
| 28 |
+
"metadata": "gguf/metadata.json",
|
| 29 |
+
"model": "gguf/model-Q4_K_M.gguf",
|
| 30 |
+
"reference": "gguf/reference.json"
|
| 31 |
+
},
|
| 32 |
+
"safetensors_f32": {
|
| 33 |
+
"config": "package/config.json",
|
| 34 |
+
"model": "package/model.safetensors",
|
| 35 |
+
"sha256": "2c696aa8fac754eebf2534453f80e67502dad635d939bf329f765b3d4f19ad2c",
|
| 36 |
+
"size_bytes": 531400
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"reference_runtime": {
|
| 40 |
+
"config_class": "LlamaConfig",
|
| 41 |
+
"implementation": "huggingface-transformers",
|
| 42 |
+
"model_class": "LlamaForCausalLM",
|
| 43 |
+
"trust_remote_code": false,
|
| 44 |
+
"version": "4.55.0"
|
| 45 |
+
},
|
| 46 |
+
"schema_version": 1,
|
| 47 |
+
"semantic_features": {
|
| 48 |
+
"explicit_head_dim": true,
|
| 49 |
+
"non_null_rope_scaling": false,
|
| 50 |
+
"query_width_matches_hidden_size": false
|
| 51 |
+
},
|
| 52 |
+
"source": {
|
| 53 |
+
"config_sha256": "e8feb37fa6f974232742d4e223e40928ccc8dabe1906ce97090e14b5753dbcc1",
|
| 54 |
+
"repository": "livekit/turn-detector",
|
| 55 |
+
"revision": "fba34c38ad5d30a63ebb83a9e6bf271cf4c91d67"
|
| 56 |
+
},
|
| 57 |
+
"validation": "POSTCHECK_OK"
|
| 58 |
+
}
|
livekit-turn-detector/config-diff.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"path": "/hidden_size",
|
| 4 |
+
"source": 576,
|
| 5 |
+
"shrunk": 48
|
| 6 |
+
},
|
| 7 |
+
{
|
| 8 |
+
"path": "/intermediate_size",
|
| 9 |
+
"source": 1536,
|
| 10 |
+
"shrunk": 96
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"path": "/max_position_embeddings",
|
| 14 |
+
"source": 8192,
|
| 15 |
+
"shrunk": 128
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"path": "/num_attention_heads",
|
| 19 |
+
"source": 9,
|
| 20 |
+
"shrunk": 6
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"path": "/num_hidden_layers",
|
| 24 |
+
"source": 30,
|
| 25 |
+
"shrunk": 2
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"path": "/num_key_value_heads",
|
| 29 |
+
"source": 3,
|
| 30 |
+
"shrunk": 2
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"path": "/vocab_size",
|
| 34 |
+
"source": 49154,
|
| 35 |
+
"shrunk": 128
|
| 36 |
+
}
|
| 37 |
+
]
|
livekit-turn-detector/gguf/metadata.json
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"commands": {
|
| 3 |
+
"convert": [
|
| 4 |
+
"PYTHON",
|
| 5 |
+
"LLAMA_CPP/convert_hf_to_gguf.py",
|
| 6 |
+
"STAGING",
|
| 7 |
+
"--outfile",
|
| 8 |
+
"INTERMEDIATE",
|
| 9 |
+
"--outtype",
|
| 10 |
+
"f16"
|
| 11 |
+
],
|
| 12 |
+
"quantize": [
|
| 13 |
+
"LLAMA_QUANTIZE",
|
| 14 |
+
"--output-tensor-type",
|
| 15 |
+
"f16",
|
| 16 |
+
"--token-embedding-type",
|
| 17 |
+
"f16",
|
| 18 |
+
"INTERMEDIATE",
|
| 19 |
+
"OUTPUT",
|
| 20 |
+
"Q4_K_M"
|
| 21 |
+
]
|
| 22 |
+
},
|
| 23 |
+
"comparison_to_f32_reference": {
|
| 24 |
+
"decode": {
|
| 25 |
+
"max_abs_error": 0.010192006826400757,
|
| 26 |
+
"mean_abs_error": 0.003219839185476303,
|
| 27 |
+
"top1_agreement": true,
|
| 28 |
+
"top5_overlap": 5
|
| 29 |
+
},
|
| 30 |
+
"prefill": {
|
| 31 |
+
"max_abs_error": 0.010380443185567856,
|
| 32 |
+
"mean_abs_error": 0.0032739201560616493,
|
| 33 |
+
"top1_agreement": true,
|
| 34 |
+
"top5_overlap": 5
|
| 35 |
+
}
|
| 36 |
+
},
|
| 37 |
+
"converter": "convert_hf_to_gguf.py",
|
| 38 |
+
"converter_log_sha256": "d80b3079bb3e6068c09523415c9b5dff069b85642acebfe10680454c2ff8f9f7",
|
| 39 |
+
"direct_token_validation": {
|
| 40 |
+
"all_logits_finite": true,
|
| 41 |
+
"decode_token_id": 13,
|
| 42 |
+
"exactly_reproducible": true,
|
| 43 |
+
"prefill_token_ids": [
|
| 44 |
+
3,
|
| 45 |
+
5,
|
| 46 |
+
7,
|
| 47 |
+
11
|
| 48 |
+
]
|
| 49 |
+
},
|
| 50 |
+
"effective_eog_token_ids": [
|
| 51 |
+
0
|
| 52 |
+
],
|
| 53 |
+
"eog_validation_range": [
|
| 54 |
+
0,
|
| 55 |
+
127
|
| 56 |
+
],
|
| 57 |
+
"format": "GGUF",
|
| 58 |
+
"intermediate_f16_sha256": "acf10d2ba51a2eb6b20248bd806115b8ea6b28ee920b0e0e9081c02ea59ac3c8",
|
| 59 |
+
"intermediate_retained": false,
|
| 60 |
+
"llama_cpp_revision": "40b740ad05c531b9d57aca6698c3ed553a9e784c",
|
| 61 |
+
"model_sha256": "29643572ec796cace456931e906b8357827a78df07007a937632bc72c4b20d0e",
|
| 62 |
+
"model_size_bytes": 211456,
|
| 63 |
+
"notes": "Q4_K_M is lossy; numerical differences are recorded as observations, not pass thresholds.",
|
| 64 |
+
"quantization_requested": "Q4_K_M",
|
| 65 |
+
"quantizer": "llama-quantize",
|
| 66 |
+
"quantizer_fallback_observed": true,
|
| 67 |
+
"quantizer_log_sha256": "134b9386c8dd87aa3986743773b5eed09cfcb57fd3cd180499fd21edcbe00d7e",
|
| 68 |
+
"schema_version": 1,
|
| 69 |
+
"source_auxiliary_tokenizer_sha256": "0f88117a2f8c52a9eabe2d61f7301b6f9b58b7a56644888da4ec9970b1efbcee",
|
| 70 |
+
"source_package_model_sha256": "2c696aa8fac754eebf2534453f80e67502dad635d939bf329f765b3d4f19ad2c",
|
| 71 |
+
"staging_special_token_normalization": {
|
| 72 |
+
"applied": false,
|
| 73 |
+
"gguf_primary_eos_token_id": 0,
|
| 74 |
+
"reason": "GGUF exposes one primary EOS metadata field; all source EOS IDs remain documented by tokenizer metadata.",
|
| 75 |
+
"source_eos_token_id": 0
|
| 76 |
+
},
|
| 77 |
+
"tensor_type_histogram": {
|
| 78 |
+
"F16": 11,
|
| 79 |
+
"F32": 5,
|
| 80 |
+
"Q5_0": 3,
|
| 81 |
+
"Q8_0": 1
|
| 82 |
+
},
|
| 83 |
+
"tiny_dimension_tensor_overrides": {
|
| 84 |
+
"output.weight": "F16",
|
| 85 |
+
"reason": "These tensors can be too narrow for the Q8_0 types selected by the Q4_K_M mixture.",
|
| 86 |
+
"token_embd.weight": "F16"
|
| 87 |
+
},
|
| 88 |
+
"tokenizer_smoke": {
|
| 89 |
+
"hf_native_ids_equal": true,
|
| 90 |
+
"text": "a",
|
| 91 |
+
"token_ids": [
|
| 92 |
+
126,
|
| 93 |
+
124
|
| 94 |
+
]
|
| 95 |
+
}
|
| 96 |
+
}
|
livekit-turn-detector/gguf/model-Q4_K_M.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:29643572ec796cace456931e906b8357827a78df07007a937632bc72c4b20d0e
|
| 3 |
+
size 211456
|
livekit-turn-detector/gguf/reference.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"model_type": "llama",
|
| 4 |
+
"file_type": 15,
|
| 5 |
+
"tokenizer_model": "llama",
|
| 6 |
+
"vocab_size": 128,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"eos_token_id": 0,
|
| 9 |
+
"pad_token_id": -1,
|
| 10 |
+
"declared_bos_token_id": 0,
|
| 11 |
+
"declared_eos_token_id": 0,
|
| 12 |
+
"declared_pad_token_id": -1,
|
| 13 |
+
"eog_token_ids": [0],
|
| 14 |
+
"prefill_token_ids": [3,5,7,11],
|
| 15 |
+
"decode_token_id": 13,
|
| 16 |
+
"smoke_text": "a",
|
| 17 |
+
"smoke_token_ids": [126,124],
|
| 18 |
+
"all_logits_finite": true,
|
| 19 |
+
"prefill_last_logits": [-0.0193141792,-0.0508461148,-0.0370915569,0.0418861434,-0.0267374776,0.0334868655,-0.0581791624,-0.106783189,0.00417452492,-0.0072201821,-0.135678887,0.393963307,0.0474759042,0.145893633,-0.00429304224,-0.0378228873,-0.0645660982,-0.130871639,0.0783228874,-0.0565563366,0.160834134,0.0292055719,-0.0419458896,-0.0695997477,0.0308215041,-0.254823059,-0.0884667635,-0.220848411,-0.0747799426,-0.00316402502,-0.221287474,0.100618228,0.00643273676,0.103701375,-0.00333619956,-0.0132760899,0.0488482974,-0.0225846898,-0.0267591011,-0.0970227495,-0.0928074047,-0.0152330138,-0.0980377197,0.013343107,0.176850468,0.111569598,-0.0780958608,-0.0073882197,0.0600401573,0.207504973,0.0816280842,-0.0153078418,0.0156859308,0.0669000968,-0.0803822502,0.00614096504,0.00623669755,-0.075031817,-0.0252885912,0.036200881,0.0508956984,0.0108062364,-0.110413782,0.074466832,0.190829247,-0.025141459,-0.063122794,-0.014329752,-0.0553358011,0.115023687,-0.0437778346,-0.161550224,0.0606371835,0.011578599,-0.0795649141,0.0762347803,-0.0815967843,0.0817829743,-0.117160194,-0.118630767,-0.0301364325,0.00722235534,-0.00664521195,-0.0988289267,-0.0765369236,0.0668949932,-0.158524439,-0.0600325055,0.00114755204,0.0832155198,-0.0583642311,0.0974971205,-0.0126464795,-0.0994286388,0.128279269,0.115928173,0.0111375647,-0.0862880573,0.105074853,-0.0555309355,0.124381274,0.0258606821,-0.154307738,-0.00299147191,-0.0472323447,0.182948828,0.134106353,0.11582496,0.0326675624,0.111265108,-0.022323722,0.112928644,0.0880587697,-0.0806296244,0.103139594,-0.153605789,0.261676699,0.0973072127,-0.0196848977,0.191981673,0.0234606713,0.09979368,0.00683066063,-0.148238093,0.0596360713,0.0776671693,-0.119186625,-0.106444068],
|
| 20 |
+
"decode_logits": [-0.0187243409,-0.00444591464,0.0557820648,0.0307359044,-0.0474938303,-0.0742083564,-0.109604962,-0.131731898,0.0352593064,0.015321577,-0.0633573085,-0.00393816689,0.126225084,0.406584442,0.0858199522,-0.00818211026,-0.102513783,-0.26781112,-0.0514992177,0.0392554067,0.11603488,0.090781346,-0.126745299,0.0193487145,-0.0126842661,-0.108849622,-0.0188514516,-0.191490501,-0.0947580114,-0.0285694525,-0.25146848,0.168008283,-0.00412976695,0.141484842,-0.0865794867,0.0226620939,0.0330909789,-0.208161429,0.098644793,-0.136219233,-0.000791996019,-0.0514821187,-0.0394399948,0.0502210967,0.0877365768,0.14125815,0.0293702371,0.0497783683,0.080882825,0.155115902,0.012048889,-0.0585217923,0.163510308,0.105159275,-0.00573973777,-0.0743498206,0.0314131267,-0.092809692,-0.0759900659,0.0648057908,-0.0778315887,0.0732942447,0.0335517228,0.0445967875,-0.00789371412,-0.075829044,-0.0245349407,-0.074946478,-0.120541088,0.131249636,-0.0169556066,-0.0799795315,0.0886956751,0.072188817,-0.046667248,-0.124494277,0.00928141642,-0.00401412044,-0.0572033189,-0.0763968155,-0.0999348685,-0.00516179902,-0.0972401723,-0.00885377172,0.0577321015,-0.0327609368,-0.0771927238,-0.0645011812,-0.0188423246,0.165469944,-0.0480766892,0.111054078,-0.141346112,0.00681997789,0.0292019937,0.130559832,0.0949311927,-0.071914047,0.124494977,0.0385251641,0.221168458,0.0506421104,-0.0209298898,0.0946206227,-0.115311213,0.172238424,0.132693172,0.135176465,-0.0800460652,0.0286448505,0.0123393442,0.0982476771,0.020662453,-0.00376544241,-0.0294984933,-0.0803721249,0.187512219,0.178083852,-0.0322346874,0.11566925,-0.0402811617,0.0256026275,-0.117471851,-0.168481424,-0.0247215908,0.0487117767,-0.0869949386,-0.0589788929]
|
| 21 |
+
}
|
livekit-turn-detector/inputs.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b49b7cca71e662452eb97b184578af4f59e9a93bc541d621986836cd7e2c395
|
| 3 |
+
size 176
|
livekit-turn-detector/package/config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_attn_implementation_autoset": true,
|
| 3 |
+
"_name_or_path": "/tmp/tmpdhw4_gdh",
|
| 4 |
+
"architectures": [
|
| 5 |
+
"LlamaForCausalLM"
|
| 6 |
+
],
|
| 7 |
+
"attention_bias": false,
|
| 8 |
+
"attention_dropout": 0.0,
|
| 9 |
+
"bos_token_id": 0,
|
| 10 |
+
"eos_token_id": 0,
|
| 11 |
+
"head_dim": 64,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 48,
|
| 14 |
+
"initializer_range": 0.041666666666666664,
|
| 15 |
+
"intermediate_size": 96,
|
| 16 |
+
"is_llama_config": true,
|
| 17 |
+
"max_position_embeddings": 128,
|
| 18 |
+
"mlp_bias": false,
|
| 19 |
+
"model_type": "llama",
|
| 20 |
+
"num_attention_heads": 6,
|
| 21 |
+
"num_hidden_layers": 2,
|
| 22 |
+
"num_key_value_heads": 2,
|
| 23 |
+
"pretraining_tp": 1,
|
| 24 |
+
"rms_norm_eps": 1e-05,
|
| 25 |
+
"rope_interleaved": false,
|
| 26 |
+
"rope_scaling": null,
|
| 27 |
+
"rope_theta": 100000,
|
| 28 |
+
"tie_word_embeddings": true,
|
| 29 |
+
"transformers_version": "4.46.3",
|
| 30 |
+
"use_cache": true,
|
| 31 |
+
"vocab_size": 128
|
| 32 |
+
}
|