Delete README.md
Browse files
README.md
DELETED
|
@@ -1,509 +0,0 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
pipeline_tag: text-generation
|
| 4 |
-
tags:
|
| 5 |
-
- fp8
|
| 6 |
-
- quantized
|
| 7 |
-
- llm-compressor
|
| 8 |
-
- compressed-tensors
|
| 9 |
-
- red hat
|
| 10 |
-
base_model:
|
| 11 |
-
- ibm-granite/granite-4.0-h-small
|
| 12 |
-
---
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
# Granite-4.0-h-small
|
| 16 |
-
|
| 17 |
-
## Model Overview
|
| 18 |
-
- **Model Architecture:** GraniteMoeHybridForCausalLM
|
| 19 |
-
- **Input:** Text
|
| 20 |
-
- **Output:** Text
|
| 21 |
-
- **Model Optimizations:**
|
| 22 |
-
- **Weight quantization:** FP8
|
| 23 |
-
- **Activation quantization:** FP8
|
| 24 |
-
- **Release Date:**
|
| 25 |
-
- **Version:** 1.0
|
| 26 |
-
- **Model Developers:**: Red Hat
|
| 27 |
-
|
| 28 |
-
Quantized version of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small).
|
| 29 |
-
|
| 30 |
-
### Model Optimizations
|
| 31 |
-
|
| 32 |
-
This model was obtained by quantizing the weights and activations of [ibm-granite/granite-4.0-h-small](https://huggingface.co/ibm-granite/granite-4.0-h-small) to FP8 data type.
|
| 33 |
-
This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
|
| 34 |
-
Only the weights and activations of the linear operators within transformers blocks of the language model are quantized.
|
| 35 |
-
|
| 36 |
-
## Deployment
|
| 37 |
-
|
| 38 |
-
### Use with vLLM
|
| 39 |
-
|
| 40 |
-
1. Initialize vLLM server:
|
| 41 |
-
```
|
| 42 |
-
vllm serve RedHatAI/granite-4.0-h-small-FP8-block --tensor_parallel_size 4
|
| 43 |
-
```
|
| 44 |
-
|
| 45 |
-
2. Send requests to the server:
|
| 46 |
-
|
| 47 |
-
```python
|
| 48 |
-
from openai import OpenAI
|
| 49 |
-
|
| 50 |
-
# Modify OpenAI's API key and API base to use vLLM's API server.
|
| 51 |
-
openai_api_key = "EMPTY"
|
| 52 |
-
openai_api_base = "http://<your-server-host>:8000/v1"
|
| 53 |
-
|
| 54 |
-
client = OpenAI(
|
| 55 |
-
api_key=openai_api_key,
|
| 56 |
-
base_url=openai_api_base,
|
| 57 |
-
)
|
| 58 |
-
|
| 59 |
-
model = "RedHatAI/granite-4.0-h-small-FP8-block"
|
| 60 |
-
|
| 61 |
-
messages = [
|
| 62 |
-
{"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
|
| 63 |
-
]
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
outputs = client.chat.completions.create(
|
| 67 |
-
model=model,
|
| 68 |
-
messages=messages,
|
| 69 |
-
)
|
| 70 |
-
|
| 71 |
-
generated_text = outputs.choices[0].message.content
|
| 72 |
-
print(generated_text)
|
| 73 |
-
```
|
| 74 |
-
|
| 75 |
-
<!-- ## Creation
|
| 76 |
-
|
| 77 |
-
This model was quantized using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as shown below.
|
| 78 |
-
|
| 79 |
-
<details>
|
| 80 |
-
<summary>Creation details</summary>
|
| 81 |
-
|
| 82 |
-
```python
|
| 83 |
-
from transformers import AutoProcessor, Qwen3ForCausalLM
|
| 84 |
-
|
| 85 |
-
from llmcompressor import oneshot
|
| 86 |
-
from llmcompressor.modeling import replace_modules_for_calibration
|
| 87 |
-
from llmcompressor.modifiers.quantization import QuantizationModifier
|
| 88 |
-
|
| 89 |
-
MODEL_ID = "Qwen/Qwen3-8B"
|
| 90 |
-
|
| 91 |
-
# Load model.
|
| 92 |
-
model = Qwen3ForCausalLM.from_pretrained(MODEL_ID, dtype="auto")
|
| 93 |
-
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
| 94 |
-
model = replace_modules_for_calibration(model)
|
| 95 |
-
|
| 96 |
-
# Configure the quantization algorithm and scheme.
|
| 97 |
-
# In this case, we:
|
| 98 |
-
# * quantize the weights to fp8 with per-block quantization
|
| 99 |
-
# * quantize the activations to fp8 with dynamic token activations
|
| 100 |
-
recipe = QuantizationModifier(
|
| 101 |
-
targets="Linear",
|
| 102 |
-
scheme="FP8_BLOCK",
|
| 103 |
-
ignore=["lm_head"],
|
| 104 |
-
)
|
| 105 |
-
|
| 106 |
-
# Apply quantization.
|
| 107 |
-
oneshot(model=model, recipe=recipe)
|
| 108 |
-
|
| 109 |
-
# Save to disk in compressed-tensors format.
|
| 110 |
-
SAVE_DIR = MODEL_ID.rstrip("/").split("/")[-1] + "-FP8-block"
|
| 111 |
-
model.save_pretrained(SAVE_DIR)
|
| 112 |
-
processor.save_pretrained(SAVE_DIR)
|
| 113 |
-
```
|
| 114 |
-
</details> -->
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
## Evaluation
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
The model was evaluated on the OpenLLM leaderboard task, using [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness).
|
| 121 |
-
[vLLM](https://docs.vllm.ai/en/stable/) was used for all evaluations.
|
| 122 |
-
|
| 123 |
-
<details>
|
| 124 |
-
<summary>Evaluation details</summary>
|
| 125 |
-
|
| 126 |
-
**Openllm V1**
|
| 127 |
-
```
|
| 128 |
-
lm_eval \
|
| 129 |
-
--model vllm \
|
| 130 |
-
--model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-block",dtype=auto,add_bos_token=True,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
|
| 131 |
-
--tasks openllm \
|
| 132 |
-
--write_out \
|
| 133 |
-
--batch_size auto \
|
| 134 |
-
--show_config
|
| 135 |
-
```
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
**Openllm V2**
|
| 139 |
-
```
|
| 140 |
-
lm_eval \
|
| 141 |
-
--model vllm \
|
| 142 |
-
--model_args pretrained="RedHatAI/granite-4.0-h-small-FP8-block",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=4,gpu_memory_utilization=0.7,disable_log_stats=True,enable_chunked_prefill=True,trust_remote_code=True \
|
| 143 |
-
--tasks leaderboard \
|
| 144 |
-
--apply_chat_template \
|
| 145 |
-
--fewshot_as_multiturn \
|
| 146 |
-
--write_out \
|
| 147 |
-
--batch_size auto \
|
| 148 |
-
--show_config
|
| 149 |
-
```
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
**Coding Benchmarks**
|
| 153 |
-
|
| 154 |
-
```
|
| 155 |
-
evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-block" \
|
| 156 |
-
--dataset "humaneval" \
|
| 157 |
-
--backend vllm \
|
| 158 |
-
--tp 4 \
|
| 159 |
-
--greedy
|
| 160 |
-
|
| 161 |
-
evalplus.evaluate --model "RedHatAI/granite-4.0-h-small-FP8-block" \
|
| 162 |
-
--dataset "mbpp" \
|
| 163 |
-
--backend vllm \
|
| 164 |
-
--tp 4 \
|
| 165 |
-
--greedy
|
| 166 |
-
|
| 167 |
-
```
|
| 168 |
-
|
| 169 |
-
</details>
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
<b>*</b> I/p Length = 2048, O/p Length = 2048, #Requests = 1024
|
| 173 |
-
|
| 174 |
-
### Accuracy
|
| 175 |
-
<table>
|
| 176 |
-
|
| 177 |
-
<thead>
|
| 178 |
-
<tr>
|
| 179 |
-
<th>Category</th>
|
| 180 |
-
<th>Metric</th>
|
| 181 |
-
<th>ibm-granite/granite-4.0-h-small</th>
|
| 182 |
-
<th>ibm-granite/granite-4.0-h-small-FP8</th>
|
| 183 |
-
<th>RedHatAI/granite-4.0-h-small-FP8-block</th>
|
| 184 |
-
</tr>
|
| 185 |
-
</thead>
|
| 186 |
-
<tbody>
|
| 187 |
-
<tr>
|
| 188 |
-
<td rowspan="1"><b>Model Size (GB)</b></td>
|
| 189 |
-
<td></td>
|
| 190 |
-
<td>64.41</td>
|
| 191 |
-
<td>33.48</td>
|
| 192 |
-
<td>36.43</td>
|
| 193 |
-
</tr>
|
| 194 |
-
<tr>
|
| 195 |
-
<td rowspan="1"><b>Throughput (Requests/sec)*</b></td>
|
| 196 |
-
<td></td>
|
| 197 |
-
<td>2.031</td>
|
| 198 |
-
<td>2.144</td>
|
| 199 |
-
<td>2.066</td>
|
| 200 |
-
</tr>
|
| 201 |
-
<tr>
|
| 202 |
-
<td rowspan="7"><b>OpenLLM V1</b></td>
|
| 203 |
-
<td>ARC-Challenge (Acc-Norm, 25-shot)</td>
|
| 204 |
-
<td>72.27</td>
|
| 205 |
-
<td>71.67 (99.17)</td>
|
| 206 |
-
<td>72.27 (100.00)</td>
|
| 207 |
-
</tr>
|
| 208 |
-
<tr>
|
| 209 |
-
<td>GSM8K (Strict-Match, 5-shot)</td>
|
| 210 |
-
<td>85.06</td>
|
| 211 |
-
<td>85.60 (100.62)</td>
|
| 212 |
-
<td>85.60 (100.62)</td>
|
| 213 |
-
</tr>
|
| 214 |
-
<tr>
|
| 215 |
-
<td>HellaSwag (Acc-Norm, 10-shot)</td>
|
| 216 |
-
<td>86.07</td>
|
| 217 |
-
<td>86.02 (99.94)</td>
|
| 218 |
-
<td>85.96 (99.87)</td>
|
| 219 |
-
</tr>
|
| 220 |
-
<tr>
|
| 221 |
-
<td>MMLU (Acc, 5-shot)</td>
|
| 222 |
-
<td>77.15</td>
|
| 223 |
-
<td>76.94 (99.73)</td>
|
| 224 |
-
<td>77.23 (100.10)</td>
|
| 225 |
-
</tr>
|
| 226 |
-
<tr>
|
| 227 |
-
<td>TruthfulQA (MC2, 0-shot)</td>
|
| 228 |
-
<td>57.97</td>
|
| 229 |
-
<td>57.62 (99.40)</td>
|
| 230 |
-
<td>57.85 (99.80)</td>
|
| 231 |
-
</tr>
|
| 232 |
-
<tr>
|
| 233 |
-
<td>Winogrande (Acc, 5-shot)</td>
|
| 234 |
-
<td>81.45</td>
|
| 235 |
-
<td>81.14 (99.61)</td>
|
| 236 |
-
<td>80.82 (99.22)</td>
|
| 237 |
-
</tr>
|
| 238 |
-
<tr>
|
| 239 |
-
<td><b>Average Score</b></td>
|
| 240 |
-
<td><b>76.66</b></td>
|
| 241 |
-
<td><b>76.50 (99.79)</b></td>
|
| 242 |
-
<td><b>76.62 (99.95)</b></td>
|
| 243 |
-
</tr>
|
| 244 |
-
<!-- OpenLLM Leaderboard V2 -->
|
| 245 |
-
<tr>
|
| 246 |
-
<td rowspan="7"><b>OpenLLM V2</b></td>
|
| 247 |
-
<td>IFEval (Inst Level Strict Acc, 0-shot)</td>
|
| 248 |
-
<td>87.41</td>
|
| 249 |
-
<td>87.65 (100.27)</td>
|
| 250 |
-
<td>87.89 (100.55)</td>
|
| 251 |
-
</tr>
|
| 252 |
-
<tr>
|
| 253 |
-
<td>BBH (Acc-Norm, 3-shot)</td>
|
| 254 |
-
<td>61.52</td>
|
| 255 |
-
<td>61.31 (99.66)</td>
|
| 256 |
-
<td>61.40 (99.80)</td>
|
| 257 |
-
</tr>
|
| 258 |
-
<tr>
|
| 259 |
-
<td>Math-Hard (Exact-Match, 4-shot)</td>
|
| 260 |
-
<td>46.60</td>
|
| 261 |
-
<td>44.34 (95.14)</td>
|
| 262 |
-
<td>44.94 (96.43)</td>
|
| 263 |
-
</tr>
|
| 264 |
-
<tr>
|
| 265 |
-
<td>GPQA (Acc-Norm, 0-shot)</td>
|
| 266 |
-
<td>32.55</td>
|
| 267 |
-
<td>32.05 (98.45)</td>
|
| 268 |
-
<td>34.23 (105.15)</td>
|
| 269 |
-
</tr>
|
| 270 |
-
<tr>
|
| 271 |
-
<td>MUSR (Acc-Norm, 0-shot)</td>
|
| 272 |
-
<td>46.43</td>
|
| 273 |
-
<td>46.30 (99.72)</td>
|
| 274 |
-
<td>45.77 (98.58)</td>
|
| 275 |
-
</tr>
|
| 276 |
-
<tr>
|
| 277 |
-
<td>MMLU-Pro (Acc, 5-shot)</td>
|
| 278 |
-
<td>47.96</td>
|
| 279 |
-
<td>47.91 (99.88)</td>
|
| 280 |
-
<td>47.93 (99.93)</td>
|
| 281 |
-
</tr>
|
| 282 |
-
<tr>
|
| 283 |
-
<td><b>Average Score</b></td>
|
| 284 |
-
<td><b>53.75</b></td>
|
| 285 |
-
<td><b>53.26 (99.09)</b></td>
|
| 286 |
-
<td><b>53.69 (99.89)</b></td>
|
| 287 |
-
</tr>
|
| 288 |
-
</tbody>
|
| 289 |
-
</table>
|
| 290 |
-
|
| 291 |
-
|
| 292 |
-
|
| 293 |
-
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
<!--
|
| 297 |
-
### Accuracy
|
| 298 |
-
<table>
|
| 299 |
-
<thead>
|
| 300 |
-
<tr>
|
| 301 |
-
<th>Category</th>
|
| 302 |
-
<th>Metric</th>
|
| 303 |
-
<th>ibm-granite/granite-4.0-h-small</th>
|
| 304 |
-
<th>ibm-granite/granite-4.0-h-small-FP8</th>
|
| 305 |
-
<th>Recovery (%)</th>
|
| 306 |
-
</tr>
|
| 307 |
-
</thead>
|
| 308 |
-
<tbody>
|
| 309 |
-
|
| 310 |
-
<tr>
|
| 311 |
-
<td rowspan="7"><b>OpenLLM V1</b></td>
|
| 312 |
-
<td>ARC-Challenge (Acc-Norm, 25-shot)</td>
|
| 313 |
-
<td>72.27</td>
|
| 314 |
-
<td>71.67</td>
|
| 315 |
-
<td>99.17</td>
|
| 316 |
-
</tr>
|
| 317 |
-
<tr>
|
| 318 |
-
<td>GSM8K (Strict-Match, 5-shot)</td>
|
| 319 |
-
<td>85.06</td>
|
| 320 |
-
<td>85.60</td>
|
| 321 |
-
<td>100.62</td>
|
| 322 |
-
</tr>
|
| 323 |
-
<tr>
|
| 324 |
-
<td>HellaSwag (Acc-Norm, 10-shot)</td>
|
| 325 |
-
<td>86.07</td>
|
| 326 |
-
<td>86.02</td>
|
| 327 |
-
<td>99.94</td>
|
| 328 |
-
</tr>
|
| 329 |
-
<tr>
|
| 330 |
-
<td>MMLU (Acc, 5-shot)</td>
|
| 331 |
-
<td>77.15</td>
|
| 332 |
-
<td>76.94</td>
|
| 333 |
-
<td>99.73</td>
|
| 334 |
-
</tr>
|
| 335 |
-
<tr>
|
| 336 |
-
<td>TruthfulQA (MC2, 0-shot)</td>
|
| 337 |
-
<td>57.97</td>
|
| 338 |
-
<td>57.62</td>
|
| 339 |
-
<td>99.40</td>
|
| 340 |
-
</tr>
|
| 341 |
-
<tr>
|
| 342 |
-
<td>Winogrande (Acc, 5-shot)</td>
|
| 343 |
-
<td>81.45</td>
|
| 344 |
-
<td>81.14</td>
|
| 345 |
-
<td>99.61</td>
|
| 346 |
-
</tr>
|
| 347 |
-
<tr>
|
| 348 |
-
<td><b>Average Score</b></td>
|
| 349 |
-
<td><b>76.66</b></td>
|
| 350 |
-
<td><b>76.50</b></td>
|
| 351 |
-
<td><b>99.79</b></td>
|
| 352 |
-
</tr>
|
| 353 |
-
|
| 354 |
-
<tr>
|
| 355 |
-
<td rowspan="7"><b>OpenLLM V2</b></td>
|
| 356 |
-
<td>IFEval (Inst Level Strict Acc, 0-shot)</td>
|
| 357 |
-
<td>87.41</td>
|
| 358 |
-
<td>87.65</td>
|
| 359 |
-
<td>100.27</td>
|
| 360 |
-
</tr>
|
| 361 |
-
<tr>
|
| 362 |
-
<td>BBH (Acc-Norm, 3-shot)</td>
|
| 363 |
-
<td>61.52</td>
|
| 364 |
-
<td>61.31</td>
|
| 365 |
-
<td>99.66</td>
|
| 366 |
-
</tr>
|
| 367 |
-
<tr>
|
| 368 |
-
<td>Math-Hard (Exact-Match, 4-shot)</td>
|
| 369 |
-
<td>46.60</td>
|
| 370 |
-
<td>44.34</td>
|
| 371 |
-
<td>95.14</td>
|
| 372 |
-
</tr>
|
| 373 |
-
<tr>
|
| 374 |
-
<td>GPQA (Acc-Norm, 0-shot)</td>
|
| 375 |
-
<td>32.55</td>
|
| 376 |
-
<td>32.05</td>
|
| 377 |
-
<td>98.45</td>
|
| 378 |
-
</tr>
|
| 379 |
-
<tr>
|
| 380 |
-
<td>MUSR (Acc-Norm, 0-shot)</td>
|
| 381 |
-
<td>46.43</td>
|
| 382 |
-
<td>46.30</td>
|
| 383 |
-
<td>99.72</td>
|
| 384 |
-
</tr>
|
| 385 |
-
<tr>
|
| 386 |
-
<td>MMLU-Pro (Acc, 5-shot)</td>
|
| 387 |
-
<td>47.96</td>
|
| 388 |
-
<td>47.91</td>
|
| 389 |
-
<td>99.88</td>
|
| 390 |
-
</tr>
|
| 391 |
-
<tr>
|
| 392 |
-
<td><b>Average Score</b></td>
|
| 393 |
-
<td><b>53.75</b></td>
|
| 394 |
-
<td><b>53.26</b></td>
|
| 395 |
-
<td><b>99.09</b></td>
|
| 396 |
-
</tr>
|
| 397 |
-
</tbody>
|
| 398 |
-
</table>
|
| 399 |
-
|
| 400 |
-
|
| 401 |
-
|
| 402 |
-
|
| 403 |
-
|
| 404 |
-
|
| 405 |
-
|
| 406 |
-
|
| 407 |
-
|
| 408 |
-
### Accuracy
|
| 409 |
-
<table>
|
| 410 |
-
<thead>
|
| 411 |
-
<tr>
|
| 412 |
-
<th>Category</th>
|
| 413 |
-
<th>Metric</th>
|
| 414 |
-
<th>ibm-granite/granite-4.0-h-small</th>
|
| 415 |
-
<th>RedHatAI/granite-4.0-h-small-FP8-block</th>
|
| 416 |
-
<th>Recovery (%)</th>
|
| 417 |
-
</tr>
|
| 418 |
-
</thead>
|
| 419 |
-
<tbody>
|
| 420 |
-
|
| 421 |
-
<tr>
|
| 422 |
-
<td rowspan="7"><b>OpenLLM V1</b></td>
|
| 423 |
-
<td>ARC-Challenge (Acc-Norm, 25-shot)</td>
|
| 424 |
-
<td>72.27</td>
|
| 425 |
-
<td>72.27</td>
|
| 426 |
-
<td>100.00</td>
|
| 427 |
-
</tr>
|
| 428 |
-
<tr>
|
| 429 |
-
<td>GSM8K (Strict-Match, 5-shot)</td>
|
| 430 |
-
<td>85.06</td>
|
| 431 |
-
<td>85.60</td>
|
| 432 |
-
<td>100.62</td>
|
| 433 |
-
</tr>
|
| 434 |
-
<tr>
|
| 435 |
-
<td>HellaSwag (Acc-Norm, 10-shot)</td>
|
| 436 |
-
<td>86.07</td>
|
| 437 |
-
<td>85.96</td>
|
| 438 |
-
<td>99.87</td>
|
| 439 |
-
</tr>
|
| 440 |
-
<tr>
|
| 441 |
-
<td>MMLU (Acc, 5-shot)</td>
|
| 442 |
-
<td>77.15</td>
|
| 443 |
-
<td>77.23</td>
|
| 444 |
-
<td>100.10</td>
|
| 445 |
-
</tr>
|
| 446 |
-
<tr>
|
| 447 |
-
<td>TruthfulQA (MC2, 0-shot)</td>
|
| 448 |
-
<td>57.97</td>
|
| 449 |
-
<td>57.85</td>
|
| 450 |
-
<td>99.80</td>
|
| 451 |
-
</tr>
|
| 452 |
-
<tr>
|
| 453 |
-
<td>Winogrande (Acc, 5-shot)</td>
|
| 454 |
-
<td>81.45</td>
|
| 455 |
-
<td>80.82</td>
|
| 456 |
-
<td>99.22</td>
|
| 457 |
-
</tr>
|
| 458 |
-
<tr>
|
| 459 |
-
<td><b>Average Score</b></td>
|
| 460 |
-
<td><b>76.66</b></td>
|
| 461 |
-
<td><b>76.62</b></td>
|
| 462 |
-
<td><b>99.95</b></td>
|
| 463 |
-
</tr>
|
| 464 |
-
|
| 465 |
-
<tr>
|
| 466 |
-
<td rowspan="7"><b>OpenLLM V2</b></td>
|
| 467 |
-
<td>IFEval (Inst Level Strict Acc, 0-shot)</td>
|
| 468 |
-
<td>87.41</td>
|
| 469 |
-
<td>87.89</td>
|
| 470 |
-
<td>100.55</td>
|
| 471 |
-
</tr>
|
| 472 |
-
<tr>
|
| 473 |
-
<td>BBH (Acc-Norm, 3-shot)</td>
|
| 474 |
-
<td>61.52</td>
|
| 475 |
-
<td>61.40</td>
|
| 476 |
-
<td>99.80</td>
|
| 477 |
-
</tr>
|
| 478 |
-
<tr>
|
| 479 |
-
<td>Math-Hard (Exact-Match, 4-shot)</td>
|
| 480 |
-
<td>46.60</td>
|
| 481 |
-
<td>44.94</td>
|
| 482 |
-
<td>96.43</td>
|
| 483 |
-
</tr>
|
| 484 |
-
<tr>
|
| 485 |
-
<td>GPQA (Acc-Norm, 0-shot)</td>
|
| 486 |
-
<td>32.55</td>
|
| 487 |
-
<td>34.23</td>
|
| 488 |
-
<td>105.15</td>
|
| 489 |
-
</tr>
|
| 490 |
-
<tr>
|
| 491 |
-
<td>MUSR (Acc-Norm, 0-shot)</td>
|
| 492 |
-
<td>46.43</td>
|
| 493 |
-
<td>45.77</td>
|
| 494 |
-
<td>98.58</td>
|
| 495 |
-
</tr>
|
| 496 |
-
<tr>
|
| 497 |
-
<td>MMLU-Pro (Acc, 5-shot)</td>
|
| 498 |
-
<td>47.96</td>
|
| 499 |
-
<td>47.93</td>
|
| 500 |
-
<td>99.93</td>
|
| 501 |
-
</tr>
|
| 502 |
-
<tr>
|
| 503 |
-
<td><b>Average Score</b></td>
|
| 504 |
-
<td><b>53.75</b></td>
|
| 505 |
-
<td><b>53.69</b></td>
|
| 506 |
-
<td><b>99.89</b></td>
|
| 507 |
-
</tr>
|
| 508 |
-
</tbody>
|
| 509 |
-
</table> -->
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|