| { |
| "models": [ |
| { |
| "model_label": "Llama-3.1-8B", |
| "model_id": "meta-llama/Llama-3.1-8B-Instruct", |
| "gpus_per_replica": 1, |
| "num_replicas": 720, |
| "initial_batch_size": 128, |
| "itl_deadline_s": 0.08, |
| "feasible_batch_sizes": [8, 16, 32, 64, 128, 256, 512] |
| }, |
| { |
| "model_label": "Llama-3.1-70B", |
| "model_id": "meta-llama/Llama-3.1-70B-Instruct", |
| "gpus_per_replica": 4, |
| "num_replicas": 180, |
| "initial_batch_size": 128, |
| "itl_deadline_s": 0.10, |
| "feasible_batch_sizes": [8, 16, 32, 64, 128, 256, 512] |
| }, |
| { |
| "model_label": "Llama-3.1-405B", |
| "model_id": "meta-llama/Llama-3.1-405B-Instruct-FP8", |
| "gpus_per_replica": 8, |
| "num_replicas": 90, |
| "initial_batch_size": 128, |
| "itl_deadline_s": 0.12, |
| "feasible_batch_sizes": [8, 16, 32, 64, 128, 256, 512] |
| }, |
| { |
| "model_label": "Qwen3-30B-A3B", |
| "model_id": "Qwen/Qwen3-30B-A3B-Thinking-2507", |
| "gpus_per_replica": 2, |
| "num_replicas": 480, |
| "initial_batch_size": 128, |
| "itl_deadline_s": 0.06, |
| "feasible_batch_sizes": [8, 16, 32, 64, 128, 256, 512] |
| }, |
| { |
| "model_label": "Qwen3-235B-A22B", |
| "model_id": "Qwen/Qwen3-235B-A22B-Thinking-2507", |
| "gpus_per_replica": 8, |
| "num_replicas": 210, |
| "initial_batch_size": 128, |
| "itl_deadline_s": 0.14, |
| "feasible_batch_sizes": [8, 16, 32, 64, 128, 256, 512] |
| } |
| ], |
| "data_sources": [ |
| {"model_label": "Llama-3.1-8B", "task": "lm-arena-chat", "gpu": "H100", "batch_sizes": [8, 16, 32, 64, 96, 128, 192, 256, 384, 512, 768, 1024]}, |
| {"model_label": "Llama-3.1-70B", "task": "lm-arena-chat", "gpu": "H100", "batch_sizes": [8, 16, 32, 64, 96, 128, 192, 256, 384, 512, 768, 1024, 1536, 2048]}, |
| {"model_label": "Llama-3.1-405B", "task": "lm-arena-chat", "gpu": "H100", "batch_sizes": [8, 16, 32, 64, 96, 128, 192, 256, 384, 512]}, |
| {"model_label": "Qwen3-30B-A3B", "task": "gpqa", "gpu": "H100", "batch_sizes": [8, 16, 32, 64, 96, 128, 192, 256, 384, 512]}, |
| {"model_label": "Qwen3-235B-A22B", "task": "gpqa", "gpu": "H100", "batch_sizes": [8, 16, 32, 64, 96, 128, 192, 256, 384, 512]} |
| ], |
| "training_trace_params": {}, |
| "data_dir": null, |
| "ieee_case_dir": "examples/ieee13", |
| "mlenergy_data_dir": null |
| } |
|
|
