Text Generation
Transformers
Safetensors
grt
geometric-rotation-transformer
mixture-of-rotations
mor
Mixture of Experts
rotational-transformer
givens-rotations
clifford-algebra
matterwave
Instructions to use MatterWaveAI/grt-30m-mor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use MatterWaveAI/grt-30m-mor with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="MatterWaveAI/grt-30m-mor")# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("MatterWaveAI/grt-30m-mor", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use MatterWaveAI/grt-30m-mor with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "MatterWaveAI/grt-30m-mor" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "MatterWaveAI/grt-30m-mor", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/MatterWaveAI/grt-30m-mor
- SGLang
How to use MatterWaveAI/grt-30m-mor with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "MatterWaveAI/grt-30m-mor" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "MatterWaveAI/grt-30m-mor", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "MatterWaveAI/grt-30m-mor" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "MatterWaveAI/grt-30m-mor", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use MatterWaveAI/grt-30m-mor with Docker Model Runner:
docker model run hf.co/MatterWaveAI/grt-30m-mor
Add 8-way classical geometric Mixture-of-Rotations optimizer ablation: eval_results.json
Browse files- eval_results.json +194 -0
eval_results.json
CHANGED
|
@@ -173,5 +173,199 @@
|
|
| 173 |
"expert_angle_l2_divergence": 2.524
|
| 174 |
}
|
| 175 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 176 |
}
|
| 177 |
}
|
|
|
|
| 173 |
"expert_angle_l2_divergence": 2.524
|
| 174 |
}
|
| 175 |
]
|
| 176 |
+
},
|
| 177 |
+
"optimizer_ablations_8way": {
|
| 178 |
+
"pair_adam": {
|
| 179 |
+
"variant": "PairAdam (Factored Toroidal Baseline)",
|
| 180 |
+
"val_loss": 0.4829,
|
| 181 |
+
"val_ppl": 1.6208,
|
| 182 |
+
"domain_val_loss": {
|
| 183 |
+
"code_and_algorithms": 0.3669,
|
| 184 |
+
"reasoning_and_language": 0.6553,
|
| 185 |
+
"rotational_geometry": 0.4772,
|
| 186 |
+
"science_and_physics": 0.5155
|
| 187 |
+
},
|
| 188 |
+
"domain_val_ppl": {
|
| 189 |
+
"code_and_algorithms": 1.4432,
|
| 190 |
+
"reasoning_and_language": 1.9257,
|
| 191 |
+
"rotational_geometry": 1.6115,
|
| 192 |
+
"science_and_physics": 1.6745
|
| 193 |
+
},
|
| 194 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 195 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 196 |
+
"mean_top1_confidence": 0.2559,
|
| 197 |
+
"expert_angle_l2_divergence": 2.5386,
|
| 198 |
+
"top4_expert_angle_l2_divergence": 2.5347,
|
| 199 |
+
"top4_full_expert_angle_l2_divergence": 3.5883,
|
| 200 |
+
"elapsed_sec": 180.15
|
| 201 |
+
},
|
| 202 |
+
"hessian_pair_adam": {
|
| 203 |
+
"variant": "HessianPairAdam (Sophia-Torus Cauchy-Riemann Curvature)",
|
| 204 |
+
"val_loss": 0.5403,
|
| 205 |
+
"val_ppl": 1.7165,
|
| 206 |
+
"domain_val_loss": {
|
| 207 |
+
"code_and_algorithms": 0.4082,
|
| 208 |
+
"reasoning_and_language": 0.7201,
|
| 209 |
+
"rotational_geometry": 0.533,
|
| 210 |
+
"science_and_physics": 0.5916
|
| 211 |
+
},
|
| 212 |
+
"domain_val_ppl": {
|
| 213 |
+
"code_and_algorithms": 1.5041,
|
| 214 |
+
"reasoning_and_language": 2.0545,
|
| 215 |
+
"rotational_geometry": 1.7041,
|
| 216 |
+
"science_and_physics": 1.8068
|
| 217 |
+
},
|
| 218 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 219 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 220 |
+
"mean_top1_confidence": 0.2559,
|
| 221 |
+
"expert_angle_l2_divergence": 2.5387,
|
| 222 |
+
"top4_expert_angle_l2_divergence": 2.5349,
|
| 223 |
+
"top4_full_expert_angle_l2_divergence": 3.5888,
|
| 224 |
+
"elapsed_sec": 180.09
|
| 225 |
+
},
|
| 226 |
+
"polar_natural_adam": {
|
| 227 |
+
"variant": "PolarNaturalAdam (Riemannian Polar Metric + Cauchy-Riemann Hessian)",
|
| 228 |
+
"val_loss": 0.4848,
|
| 229 |
+
"val_ppl": 1.6239,
|
| 230 |
+
"domain_val_loss": {
|
| 231 |
+
"code_and_algorithms": 0.3744,
|
| 232 |
+
"reasoning_and_language": 0.6466,
|
| 233 |
+
"rotational_geometry": 0.4747,
|
| 234 |
+
"science_and_physics": 0.5226
|
| 235 |
+
},
|
| 236 |
+
"domain_val_ppl": {
|
| 237 |
+
"code_and_algorithms": 1.4542,
|
| 238 |
+
"reasoning_and_language": 1.9091,
|
| 239 |
+
"rotational_geometry": 1.6076,
|
| 240 |
+
"science_and_physics": 1.6865
|
| 241 |
+
},
|
| 242 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 243 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 244 |
+
"mean_top1_confidence": 0.2559,
|
| 245 |
+
"expert_angle_l2_divergence": 2.5387,
|
| 246 |
+
"top4_expert_angle_l2_divergence": 2.5349,
|
| 247 |
+
"top4_full_expert_angle_l2_divergence": 3.5887,
|
| 248 |
+
"elapsed_sec": 180.75
|
| 249 |
+
},
|
| 250 |
+
"cyclide_adam": {
|
| 251 |
+
"variant": "CyclideAdam (Dupin Cyclide Conic Metric + Deviatoric Twist)",
|
| 252 |
+
"val_loss": 0.5006,
|
| 253 |
+
"val_ppl": 1.6498,
|
| 254 |
+
"domain_val_loss": {
|
| 255 |
+
"code_and_algorithms": 0.3758,
|
| 256 |
+
"reasoning_and_language": 0.6718,
|
| 257 |
+
"rotational_geometry": 0.4986,
|
| 258 |
+
"science_and_physics": 0.5426
|
| 259 |
+
},
|
| 260 |
+
"domain_val_ppl": {
|
| 261 |
+
"code_and_algorithms": 1.4561,
|
| 262 |
+
"reasoning_and_language": 1.9578,
|
| 263 |
+
"rotational_geometry": 1.6464,
|
| 264 |
+
"science_and_physics": 1.7205
|
| 265 |
+
},
|
| 266 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 267 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 268 |
+
"mean_top1_confidence": 0.2559,
|
| 269 |
+
"expert_angle_l2_divergence": 2.5534,
|
| 270 |
+
"top4_expert_angle_l2_divergence": 2.5789,
|
| 271 |
+
"top4_full_expert_angle_l2_divergence": 3.6659,
|
| 272 |
+
"elapsed_sec": 180.09
|
| 273 |
+
},
|
| 274 |
+
"symplectic_adam": {
|
| 275 |
+
"variant": "SymplecticAdam (Gromov-Liouville Non-Squeezing Floor + Casimir Cavity)",
|
| 276 |
+
"val_loss": 0.5567,
|
| 277 |
+
"val_ppl": 1.7448,
|
| 278 |
+
"domain_val_loss": {
|
| 279 |
+
"code_and_algorithms": 0.4185,
|
| 280 |
+
"reasoning_and_language": 0.7417,
|
| 281 |
+
"rotational_geometry": 0.553,
|
| 282 |
+
"science_and_physics": 0.6082
|
| 283 |
+
},
|
| 284 |
+
"domain_val_ppl": {
|
| 285 |
+
"code_and_algorithms": 1.5197,
|
| 286 |
+
"reasoning_and_language": 2.0996,
|
| 287 |
+
"rotational_geometry": 1.7385,
|
| 288 |
+
"science_and_physics": 1.8372
|
| 289 |
+
},
|
| 290 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 291 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 292 |
+
"mean_top1_confidence": 0.2559,
|
| 293 |
+
"expert_angle_l2_divergence": 2.5386,
|
| 294 |
+
"top4_expert_angle_l2_divergence": 2.5345,
|
| 295 |
+
"top4_full_expert_angle_l2_divergence": 3.588,
|
| 296 |
+
"elapsed_sec": 180.2
|
| 297 |
+
},
|
| 298 |
+
"gyric_adam": {
|
| 299 |
+
"variant": "GyricAdam (Poinsot-Lande Gyroscopic Projection + Thomas Precession)",
|
| 300 |
+
"val_loss": 0.5745,
|
| 301 |
+
"val_ppl": 1.7763,
|
| 302 |
+
"domain_val_loss": {
|
| 303 |
+
"code_and_algorithms": 0.4279,
|
| 304 |
+
"reasoning_and_language": 0.7609,
|
| 305 |
+
"rotational_geometry": 0.5672,
|
| 306 |
+
"science_and_physics": 0.6412
|
| 307 |
+
},
|
| 308 |
+
"domain_val_ppl": {
|
| 309 |
+
"code_and_algorithms": 1.5341,
|
| 310 |
+
"reasoning_and_language": 2.1401,
|
| 311 |
+
"rotational_geometry": 1.7633,
|
| 312 |
+
"science_and_physics": 1.8987
|
| 313 |
+
},
|
| 314 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 315 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 316 |
+
"mean_top1_confidence": 0.2559,
|
| 317 |
+
"expert_angle_l2_divergence": 2.5387,
|
| 318 |
+
"top4_expert_angle_l2_divergence": 2.535,
|
| 319 |
+
"top4_full_expert_angle_l2_divergence": 3.5888,
|
| 320 |
+
"elapsed_sec": 177.29
|
| 321 |
+
},
|
| 322 |
+
"huygens_adam": {
|
| 323 |
+
"variant": "HuygensAdam (Circular Order Parameter R_p Splay Torque)",
|
| 324 |
+
"val_loss": 0.5225,
|
| 325 |
+
"val_ppl": 1.6862,
|
| 326 |
+
"domain_val_loss": {
|
| 327 |
+
"code_and_algorithms": 0.3957,
|
| 328 |
+
"reasoning_and_language": 0.7041,
|
| 329 |
+
"rotational_geometry": 0.5107,
|
| 330 |
+
"science_and_physics": 0.5696
|
| 331 |
+
},
|
| 332 |
+
"domain_val_ppl": {
|
| 333 |
+
"code_and_algorithms": 1.4854,
|
| 334 |
+
"reasoning_and_language": 2.022,
|
| 335 |
+
"rotational_geometry": 1.6665,
|
| 336 |
+
"science_and_physics": 1.7676
|
| 337 |
+
},
|
| 338 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 339 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 340 |
+
"mean_top1_confidence": 0.2559,
|
| 341 |
+
"expert_angle_l2_divergence": 2.5676,
|
| 342 |
+
"top4_expert_angle_l2_divergence": 2.6216,
|
| 343 |
+
"top4_full_expert_angle_l2_divergence": 3.7258,
|
| 344 |
+
"elapsed_sec": 176.07
|
| 345 |
+
},
|
| 346 |
+
"huygens_symplectic_adam": {
|
| 347 |
+
"variant": "HuygensSymplecticAdam (Two-Stage Huygens -> Symplectic Curriculum)",
|
| 348 |
+
"val_loss": 0.523,
|
| 349 |
+
"val_ppl": 1.6871,
|
| 350 |
+
"domain_val_loss": {
|
| 351 |
+
"code_and_algorithms": 0.4091,
|
| 352 |
+
"reasoning_and_language": 0.6859,
|
| 353 |
+
"rotational_geometry": 0.5202,
|
| 354 |
+
"science_and_physics": 0.557
|
| 355 |
+
},
|
| 356 |
+
"domain_val_ppl": {
|
| 357 |
+
"code_and_algorithms": 1.5055,
|
| 358 |
+
"reasoning_and_language": 1.9856,
|
| 359 |
+
"rotational_geometry": 1.6823,
|
| 360 |
+
"science_and_physics": 1.7454
|
| 361 |
+
},
|
| 362 |
+
"mean_routing_entropy_nats": 1.3862,
|
| 363 |
+
"mean_routing_entropy_bits": 1.9999,
|
| 364 |
+
"mean_top1_confidence": 0.2559,
|
| 365 |
+
"expert_angle_l2_divergence": 2.5388,
|
| 366 |
+
"top4_expert_angle_l2_divergence": 2.5351,
|
| 367 |
+
"top4_full_expert_angle_l2_divergence": 3.5896,
|
| 368 |
+
"elapsed_sec": 176.01
|
| 369 |
+
}
|
| 370 |
}
|
| 371 |
}
|