Borrison commited on
Commit
d626d1b
·
verified ·
1 Parent(s): 072d822

Delete args.json

Browse files
Files changed (1) hide show
  1. args.json +0 -784
args.json DELETED
@@ -1,784 +0,0 @@
1
- {
2
- "use_ray": false,
3
- "ray_exp_name": null,
4
- "device_groups": null,
5
- "model": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900",
6
- "model_type": "qwen3_omni_moe",
7
- "model_revision": null,
8
- "task_type": "causal_lm",
9
- "torch_dtype": "bfloat16",
10
- "attn_impl": null,
11
- "experts_impl": null,
12
- "new_special_tokens": [],
13
- "num_labels": null,
14
- "problem_type": null,
15
- "rope_scaling": {
16
- "interleaved": true,
17
- "mrope_interleaved": true,
18
- "mrope_section": [
19
- 24,
20
- 20,
21
- 20
22
- ],
23
- "rope_type": "default",
24
- "type": "default"
25
- },
26
- "device_map": null,
27
- "max_memory": {},
28
- "max_model_len": null,
29
- "local_repo_path": null,
30
- "init_strategy": null,
31
- "template": "qwen3_omni",
32
- "system": "",
33
- "max_length": 32768,
34
- "truncation_strategy": "left",
35
- "max_pixels": null,
36
- "agent_template": null,
37
- "norm_bbox": null,
38
- "use_chat_template": true,
39
- "padding_side": "right",
40
- "padding_free": true,
41
- "loss_scale": "last_round",
42
- "sequence_parallel_size": 1,
43
- "template_backend": "swift",
44
- "response_prefix": null,
45
- "enable_thinking": null,
46
- "add_non_thinking_prefix": true,
47
- "dataset": [
48
- "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/data/raw/stage2/20260123_merged_shuffled_filtered_qy.jsonl"
49
- ],
50
- "val_dataset": [],
51
- "cached_dataset": [],
52
- "cached_val_dataset": [],
53
- "split_dataset_ratio": 0.0,
54
- "data_seed": 42,
55
- "dataset_num_proc": 2,
56
- "load_from_cache_file": true,
57
- "dataset_shuffle": true,
58
- "val_dataset_shuffle": false,
59
- "streaming": false,
60
- "interleave_prob": null,
61
- "stopping_strategy": "first_exhausted",
62
- "shuffle_buffer_size": 1000,
63
- "download_mode": "reuse_dataset_if_exists",
64
- "columns": {},
65
- "strict": false,
66
- "remove_unused_columns": false,
67
- "model_name": null,
68
- "model_author": null,
69
- "custom_dataset_info": [],
70
- "quant_method": null,
71
- "quant_bits": null,
72
- "hqq_axis": null,
73
- "bnb_4bit_compute_dtype": "bfloat16",
74
- "bnb_4bit_quant_type": "nf4",
75
- "bnb_4bit_use_double_quant": true,
76
- "bnb_4bit_quant_storage": null,
77
- "max_new_tokens": null,
78
- "temperature": 1.2,
79
- "top_k": 50,
80
- "top_p": 0.99,
81
- "repetition_penalty": 1.0,
82
- "num_beams": 1,
83
- "stream": false,
84
- "stop_words": [],
85
- "logprobs": false,
86
- "top_logprobs": null,
87
- "structured_outputs_regex": null,
88
- "tuner_backend": "peft",
89
- "tuner_type": "full",
90
- "train_type": null,
91
- "adapters": [],
92
- "external_plugins": [
93
- "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/src/examples/multi_node/grpo/plugin_think_plain.py"
94
- ],
95
- "custom_register_path": [],
96
- "seed": 42,
97
- "model_kwargs": {},
98
- "load_args": false,
99
- "load_data_args": false,
100
- "packing": false,
101
- "packing_length": null,
102
- "packing_num_proc": 1,
103
- "lazy_tokenize": true,
104
- "use_hf": false,
105
- "hub_token": null,
106
- "ddp_timeout": 18000000,
107
- "ddp_backend": null,
108
- "ignore_args_error": false,
109
- "use_swift_lora": false,
110
- "freeze_llm": false,
111
- "freeze_vit": false,
112
- "freeze_aligner": false,
113
- "freeze_parameters": [],
114
- "freeze_parameters_regex": null,
115
- "freeze_parameters_ratio": 0.0,
116
- "trainable_parameters": [],
117
- "trainable_parameters_regex": null,
118
- "adapter_load": null,
119
- "target_modules": [
120
- "all-linear"
121
- ],
122
- "target_regex": null,
123
- "modules_to_save": [],
124
- "lora_rank": 128,
125
- "lora_alpha": 512,
126
- "lora_dropout": 0.05,
127
- "lora_bias": "none",
128
- "lora_dtype": null,
129
- "use_rslora": false,
130
- "rlhf_type": "mapo",
131
- "ref_load": null,
132
- "ref_adapter_load": null,
133
- "beta": 0.001,
134
- "rpo_alpha": null,
135
- "reference_free": false,
136
- "label_smoothing": 0.0,
137
- "f_divergence_type": "reverse_kl",
138
- "desirable_weight": 1.0,
139
- "undesirable_weight": 1.0,
140
- "calculate_KL": null,
141
- "center_rewards_coefficient": null,
142
- "teacher_model": null,
143
- "teacher_model_type": null,
144
- "teacher_model_revision": null,
145
- "lmbda": 0.5,
146
- "seq_kd": false,
147
- "offload_teacher_model": false,
148
- "sft_alpha": 0.0,
149
- "generation_batch_size": 768,
150
- "steps_per_generation": 4,
151
- "num_generations": 8,
152
- "num_generations_eval": null,
153
- "max_completion_length": 1024,
154
- "importance_sampling_level": "token",
155
- "tau_pos": 1.0,
156
- "tau_neg": 1.05,
157
- "epsilon": 0.2,
158
- "epsilon_high": null,
159
- "delta": null,
160
- "use_vllm": true,
161
- "vllm_mode": "server",
162
- "vllm_enable_prefix_caching": true,
163
- "vllm_gpu_memory_utilization": 0.9,
164
- "vllm_tensor_parallel_size": 1,
165
- "vllm_max_model_len": null,
166
- "vllm_enforce_eager": false,
167
- "vllm_limit_mm_per_prompt": null,
168
- "vllm_disable_cascade_attn": false,
169
- "vllm_max_num_seqs": null,
170
- "vllm_mm_processor_cache_gb": null,
171
- "vllm_engine_kwargs": {},
172
- "sleep_level": 0,
173
- "offload_optimizer": false,
174
- "offload_model": true,
175
- "offload_bridge": false,
176
- "vllm_server_base_url": null,
177
- "vllm_server_host": [
178
- "29.164.184.102"
179
- ],
180
- "vllm_server_port": [
181
- 11086
182
- ],
183
- "vllm_server_timeout": 240.0,
184
- "vllm_server_group_port": [
185
- 51313
186
- ],
187
- "reward_funcs": [
188
- "external_mcqa_accuracy",
189
- "external_format",
190
- "external_consistency"
191
- ],
192
- "reward_weights": [
193
- 2.0,
194
- 1.0,
195
- 1.0
196
- ],
197
- "cosine_min_len_value_wrong": -0.5,
198
- "cosine_max_len_value_wrong": 0.0,
199
- "cosine_min_len_value_correct": 1.0,
200
- "cosine_max_len_value_correct": 0.5,
201
- "cosine_max_len": null,
202
- "repetition_n_grams": 3,
203
- "repetition_max_penalty": -1.0,
204
- "soft_max_length": null,
205
- "soft_cache_length": null,
206
- "dynamic_sample": false,
207
- "max_resample_times": 3,
208
- "overlong_filter": true,
209
- "scale_rewards": "group",
210
- "advantage_estimator": "grpo",
211
- "kl_in_reward": false,
212
- "wandb_log_unique_prompts": null,
213
- "log_completions": true,
214
- "rollout_importance_sampling_mode": null,
215
- "rollout_importance_sampling_threshold": 2.0,
216
- "log_rollout_offpolicy_metrics": false,
217
- "off_policy_sequence_mask_delta": null,
218
- "log_entropy": false,
219
- "top_entropy_quantile": 1.0,
220
- "eta": 0.5,
221
- "mapo_advantage_floor_eps": 0.1,
222
- "mapo_task_fail_gate_floor": 0.0,
223
- "mapo_mask_temperature": 1.0,
224
- "mapo_temporal_kappa": 1.0,
225
- "mapo_attention_layers": "43,44,45,46,47",
226
- "mapo_attention_head_reduce": "max",
227
- "mapo_attention_layer_reduce": "mean",
228
- "mapo_failure_reward_name": "MCQAAccuracy",
229
- "mapo_failure_threshold": 0.0,
230
- "mapo_pos_tags": "NOUN,VERB,ADJ,ADV,NUM,X",
231
- "mapo_attention_only": false,
232
- "mapo_debug_attn_grad_probe": false,
233
- "mapo_debug_attn_grad_probe_interval": 0,
234
- "diff_objective": "softplus_margin",
235
- "diff_kl_type": "forward",
236
- "diff_margin_gamma": 2.0,
237
- "diff_margin_beta": 4.0,
238
- "entropy_mask_quantile": 0.25,
239
- "entropy_mask_scope": "sequence",
240
- "entropy_mask_type": "hard",
241
- "text_only_modality_scope": "audio",
242
- "text_ref_model": null,
243
- "text_ref_load": null,
244
- "reward_model": null,
245
- "reward_model_plugin": null,
246
- "sync_ref_model": false,
247
- "ref_model_sync_steps": 512,
248
- "ref_model_mixup_alpha": 0.6,
249
- "async_generate": false,
250
- "move_model_batches": null,
251
- "multi_turn_scheduler": null,
252
- "max_turns": null,
253
- "completion_length_limit_scope": "per_round",
254
- "vllm_server_pass_dataset": false,
255
- "num_iterations": 1,
256
- "loss_type": "grpo",
257
- "check_model": true,
258
- "padded_vocab_size": 152064,
259
- "initialize_embedding": false,
260
- "mlp_padding_free": false,
261
- "load_safetensors": true,
262
- "save_safetensors": true,
263
- "ref_model": null,
264
- "ref_adapters": [],
265
- "merge_lora": false,
266
- "max_shard_size": "5GB",
267
- "train_dataloader_shuffle": true,
268
- "dataloader_pin_memory": true,
269
- "dataloader_persistent_workers": true,
270
- "dataloader_prefetch_factor": 2,
271
- "group_by_length": false,
272
- "hf_model_type": "qwen3_omni_moe",
273
- "llm_model_type": "qwen3_omni_moe_text",
274
- "max_epochs": null,
275
- "enable_dft_loss": false,
276
- "enable_channel_loss": false,
277
- "save_strategy": "steps",
278
- "original_max_position_embeddings": null,
279
- "partial_rotary_factor": null,
280
- "use_shared_expert_gate": false,
281
- "report_to": null,
282
- "vit_gradient_checkpointing": true,
283
- "vit_lr": null,
284
- "aligner_lr": null,
285
- "gradient_checkpointing_kwargs": null,
286
- "linear_num_value_heads": null,
287
- "linear_num_key_heads": null,
288
- "linear_key_head_dim": null,
289
- "linear_value_head_dim": null,
290
- "linear_conv_kernel_dim": null,
291
- "layer_types": null,
292
- "apply_wd_to_qk_layernorm": false,
293
- "mrope_interleaved": true,
294
- "micro_batch_size": 4,
295
- "global_batch_size": 192,
296
- "recompute_granularity": "selective",
297
- "recompute_method": null,
298
- "recompute_num_layers": null,
299
- "recompute_modules": [
300
- "core_attn"
301
- ],
302
- "use_cpu_initialization": false,
303
- "deterministic_mode": false,
304
- "train_iters": 1000,
305
- "log_interval": 1,
306
- "tensorboard_dir": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/checkpoint-900/v5-20260406-174200/runs",
307
- "no_masked_softmax_fusion": false,
308
- "no_bias_dropout_fusion": false,
309
- "no_bias_swiglu_fusion": false,
310
- "no_rope_fusion": false,
311
- "no_gradient_accumulation_fusion": true,
312
- "cross_entropy_loss_fusion": false,
313
- "cross_entropy_fusion_impl": "native",
314
- "calculate_per_token_loss": false,
315
- "use_flash_attn": false,
316
- "attention_backend": "flash",
317
- "optimizer": "adam",
318
- "optimizer_cpu_offload": true,
319
- "optimizer_offload_fraction": 1.0,
320
- "use_precision_aware_optimizer": true,
321
- "main_grads_dtype": "fp32",
322
- "main_params_dtype": "fp32",
323
- "exp_avg_dtype": "fp32",
324
- "exp_avg_sq_dtype": "fp32",
325
- "dataloader_type": "cyclic",
326
- "manual_gc": false,
327
- "manual_gc_interval": 0,
328
- "lr": 1e-06,
329
- "lr_decay_style": "cosine",
330
- "lr_decay_iters": null,
331
- "lr_warmup_iters": 0,
332
- "lr_warmup_fraction": 0.05,
333
- "min_lr": 0,
334
- "weight_decay": 0.1,
335
- "clip_grad": 1.0,
336
- "adam_beta1": 0.9,
337
- "adam_beta2": 0.95,
338
- "adam_eps": 1e-08,
339
- "sgd_momentum": 0.9,
340
- "save": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/checkpoint-900/v5-20260406-174200",
341
- "save_interval": 100,
342
- "save_retain_interval": null,
343
- "no_save_optim": true,
344
- "no_save_rng": true,
345
- "load": null,
346
- "no_load_optim": false,
347
- "no_load_rng": false,
348
- "finetune": true,
349
- "ckpt_format": "torch_dist",
350
- "no_initialization": true,
351
- "auto_detect_ckpt_format": true,
352
- "exit_on_missing_checkpoint": true,
353
- "async_save": false,
354
- "use_persistent_ckpt_worker": false,
355
- "ckpt_fully_parallel_load": false,
356
- "ckpt_assume_constant_structure": false,
357
- "distributed_backend": "nccl",
358
- "local_rank": 0,
359
- "use_distributed_optimizer": true,
360
- "tensor_model_parallel_size": 4,
361
- "pipeline_model_parallel_size": 2,
362
- "decoder_first_pipeline_num_layers": null,
363
- "decoder_last_pipeline_num_layers": null,
364
- "account_for_embedding_in_pipeline_split": false,
365
- "account_for_loss_in_pipeline_split": false,
366
- "sequence_parallel": true,
367
- "context_parallel_size": 1,
368
- "tp_comm_overlap": false,
369
- "overlap_grad_reduce": false,
370
- "overlap_param_gather": false,
371
- "distributed_timeout_minutes": 300000,
372
- "num_layers_per_virtual_pipeline_stage": null,
373
- "num_virtual_stages_per_pipeline_rank": null,
374
- "microbatch_group_size_per_virtual_pipeline_stage": null,
375
- "pipeline_model_parallel_layout": null,
376
- "num_layers": 48,
377
- "hidden_size": 2048,
378
- "ffn_hidden_size": null,
379
- "num_attention_heads": 32,
380
- "group_query_attention": true,
381
- "num_query_groups": 4,
382
- "softmax_type": "vanilla",
383
- "window_size": null,
384
- "window_attn_skip_freq": null,
385
- "max_position_embeddings": 65536,
386
- "position_embedding_type": "mrope",
387
- "mrope_section": [
388
- 24,
389
- 20,
390
- 20
391
- ],
392
- "rotary_base": 1000000,
393
- "rotary_percent": 1.0,
394
- "rotary_interleaved": false,
395
- "normalization": "RMSNorm",
396
- "norm_epsilon": 1e-06,
397
- "swiglu": true,
398
- "quick_geglu": false,
399
- "activation_func_clamp_value": null,
400
- "glu_linear_offset": 0.0,
401
- "untie_embeddings_and_output_weights": true,
402
- "disable_bias_linear": true,
403
- "add_qkv_bias": false,
404
- "attention_dropout": 0.0,
405
- "hidden_dropout": 0.0,
406
- "kv_channels": 128,
407
- "qk_layernorm": true,
408
- "qk_l2_norm": null,
409
- "no_rope_freq": null,
410
- "moe_apply_probs_on_input": null,
411
- "transformer_impl": "transformer_engine",
412
- "num_experts": 128,
413
- "moe_layer_freq": 1,
414
- "moe_ffn_hidden_size": 768,
415
- "moe_shared_expert_intermediate_size": null,
416
- "moe_router_topk": 8,
417
- "moe_router_num_groups": null,
418
- "moe_router_group_topk": null,
419
- "moe_router_pre_softmax": false,
420
- "moe_router_dtype": "fp32",
421
- "moe_router_score_function": "softmax",
422
- "moe_router_bias_update_rate": null,
423
- "moe_router_enable_expert_bias": false,
424
- "moe_router_topk_scaling_factor": null,
425
- "moe_router_load_balancing_type": "aux_loss",
426
- "expert_model_parallel_size": 4,
427
- "expert_tensor_parallel_size": 1,
428
- "moe_token_dispatcher_type": null,
429
- "moe_enable_deepep": false,
430
- "moe_grouped_gemm": true,
431
- "moe_permute_fusion": true,
432
- "moe_aux_loss_coeff": 0.0001,
433
- "moe_z_loss_coeff": null,
434
- "moe_shared_expert_overlap": true,
435
- "moe_layer_recompute": false,
436
- "moe_expert_capacity_factor": null,
437
- "moe_pad_expert_input_to_capacity": false,
438
- "moe_token_drop_policy": null,
439
- "multi_latent_attention": false,
440
- "q_lora_rank": null,
441
- "kv_lora_rank": 32,
442
- "qk_head_dim": 128,
443
- "qk_pos_emb_head_dim": 64,
444
- "v_head_dim": 128,
445
- "mtp_num_layers": null,
446
- "mtp_loss_scaling_factor": 0.1,
447
- "fp8_format": null,
448
- "fp8_recipe": "delayed",
449
- "fp8_amax_history_len": 1024,
450
- "fp8_amax_compute_algo": "max",
451
- "fp8_param_gather": false,
452
- "fp16": false,
453
- "bf16": true,
454
- "apply_query_key_layer_scaling": false,
455
- "attention_softmax_in_fp32": true,
456
- "log_params_norm": false,
457
- "log_throughput": false,
458
- "tensorboard_log_interval": 1,
459
- "tensorboard_queue_size": 50,
460
- "log_timers_to_tensorboard": true,
461
- "no_log_learning_rate_to_tensorboard": false,
462
- "log_validation_ppl_to_tensorboard": true,
463
- "log_memory_to_tensorboard": true,
464
- "logging_level": null,
465
- "wandb_project": "megatron-swift",
466
- "wandb_exp_name": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/checkpoint-900/v5-20260406-174200",
467
- "wandb_save_dir": null,
468
- "eval_iters": -1,
469
- "eval_interval": 200,
470
- "seq_length": 32768,
471
- "num_workers": 2,
472
- "no_data_sharding": false,
473
- "megatron_extra_kwargs": {},
474
- "add_version": true,
475
- "swift_version": "4.0.0.dev0",
476
- "ckpt_dir": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900",
477
- "rank": 0,
478
- "global_world_size": 24,
479
- "local_world_size": 8,
480
- "model_suffix": "checkpoint-900",
481
- "model_info": "ModelInfo(model_type='qwen3_omni_moe', model_dir='/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900', torch_dtype=torch.bfloat16, max_model_len=32768, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=True, is_multimodal=True, config=None, task_type='causal_lm', num_labels=None)",
482
- "model_meta": "ModelMeta(model_type='qwen3_omni_moe', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-Omni-30B-A3B-Instruct', hf_model_id='Qwen/Qwen3-Omni-30B-A3B-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-Omni-30B-A3B-Thinking', hf_model_id='Qwen/Qwen3-Omni-30B-A3B-Thinking', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-Omni-30B-A3B-Captioner', hf_model_id='Qwen/Qwen3-Omni-30B-A3B-Captioner', model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_omni', ignore_patterns=None, requires=None, tags=[])], loader=<class 'swift.model.models.qwen.Qwen3OmniLoader'>, template=None, model_arch=MultiModelKeys(arch_name='qwen3_omni', embedding=None, module_list=None, lm_head=None, q_proj=None, k_proj=None, v_proj=None, o_proj=None, attention=None, mlp=None, down_proj=None, qkv_proj=None, qk_proj=None, qa_proj=None, qb_proj=None, kv_proj=None, kva_proj=None, kvb_proj=None, language_model=['thinker.model', 'thinker.lm_head'], aligner=['thinker.audio_tower.proj1', 'thinker.audio_tower.proj2', 'thinker.visual.merger', 'thinker.visual.merger_list'], vision_tower=['thinker.audio_tower', 'thinker.visual'], generator=['talker', 'code2wav']), architectures=['Qwen3OmniMoeForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=['transformers>=4.57.dev0', 'soundfile', 'decord', 'qwen_omni_utils'], tags=['vision', 'video', 'audio'])",
483
- "model_dir": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900",
484
- "template_meta": "QwenTemplateMeta(template_type='qwen3_omni', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'swift.template.templates.qwen.Qwen3OmniTemplate'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=False, thinking_prefix='<think>\\n', non_thinking_prefix='', history_thinking_prefix='')",
485
- "_val_dataset_exists": [],
486
- "hub": "<class 'swift.hub.hub.MSHub'>",
487
- "megatron_model_meta": "MegatronModelMeta(megatron_model_type='qwen3_omni', model_types=['qwen3_omni_moe'], is_multimodal=True, bridge_cls=<class 'swift.megatron.model.mm_gpts.qwen3_vl.Qwen3OmniBridge'>, model_cls=<class 'swift.megatron.model.mm_gpts.qwen3_vl.Qwen3VLGPTModel'>, get_transformer_layer_spec=None, model_provider=<function model_provider at 0x7fee5e3642c0>, visual_cls=<class 'swift.megatron.model.mm_gpts.qwen3_vl.Qwen3Omni_Vit'>, get_mtp_block_spec=None, extra_args_provider=None)",
488
- "per_device_generation_batch_size": 32,
489
- "extra_args": {
490
- "model_dir": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900",
491
- "is_multimodal": true,
492
- "hf_model_type": "qwen3_omni_moe",
493
- "use_ray": false,
494
- "ray_exp_name": null,
495
- "device_groups": null,
496
- "model": "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/exp/mapo/Qwen3-Omni-30B-A3B-Thinking/v35-20260403-165326/checkpoint-900",
497
- "model_type": "qwen3_omni_moe",
498
- "model_revision": null,
499
- "task_type": "causal_lm",
500
- "torch_dtype": "bfloat16",
501
- "attn_impl": null,
502
- "experts_impl": null,
503
- "new_special_tokens": [],
504
- "num_labels": null,
505
- "problem_type": null,
506
- "rope_scaling": {
507
- "interleaved": true,
508
- "mrope_interleaved": true,
509
- "mrope_section": [
510
- 24,
511
- 20,
512
- 20
513
- ],
514
- "rope_type": "default",
515
- "type": "default"
516
- },
517
- "device_map": null,
518
- "max_memory": {},
519
- "max_model_len": null,
520
- "local_repo_path": null,
521
- "init_strategy": null,
522
- "template": "qwen3_omni",
523
- "system": "",
524
- "max_length": 32768,
525
- "truncation_strategy": "left",
526
- "max_pixels": null,
527
- "agent_template": null,
528
- "norm_bbox": null,
529
- "use_chat_template": true,
530
- "padding_side": "right",
531
- "padding_free": true,
532
- "sequence_parallel_size": 1,
533
- "template_backend": "swift",
534
- "response_prefix": null,
535
- "enable_thinking": null,
536
- "add_non_thinking_prefix": true,
537
- "dataset": [
538
- "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/data/raw/stage2/20260123_merged_shuffled_filtered_qy.jsonl"
539
- ],
540
- "val_dataset": [],
541
- "cached_dataset": [],
542
- "cached_val_dataset": [],
543
- "split_dataset_ratio": 0.0,
544
- "data_seed": 42,
545
- "dataset_num_proc": 2,
546
- "load_from_cache_file": true,
547
- "dataset_shuffle": true,
548
- "val_dataset_shuffle": false,
549
- "streaming": false,
550
- "interleave_prob": null,
551
- "stopping_strategy": "first_exhausted",
552
- "shuffle_buffer_size": 1000,
553
- "download_mode": "reuse_dataset_if_exists",
554
- "columns": {},
555
- "strict": false,
556
- "remove_unused_columns": false,
557
- "model_name": null,
558
- "model_author": null,
559
- "custom_dataset_info": [],
560
- "quant_method": null,
561
- "quant_bits": null,
562
- "hqq_axis": null,
563
- "bnb_4bit_compute_dtype": "bfloat16",
564
- "bnb_4bit_quant_type": "nf4",
565
- "bnb_4bit_use_double_quant": true,
566
- "bnb_4bit_quant_storage": null,
567
- "max_new_tokens": null,
568
- "temperature": 1.2,
569
- "top_k": 50,
570
- "top_p": 0.99,
571
- "repetition_penalty": 1.0,
572
- "num_beams": 1,
573
- "stream": false,
574
- "stop_words": [],
575
- "logprobs": false,
576
- "top_logprobs": null,
577
- "structured_outputs_regex": null,
578
- "tuner_backend": "peft",
579
- "tuner_type": "full",
580
- "train_type": null,
581
- "adapters": [],
582
- "external_plugins": [
583
- "/apdcephfs_qy3/share_302401520/user/cihanxiao/cihanxiao/mapo/src/examples/multi_node/grpo/plugin_think_plain.py"
584
- ],
585
- "custom_register_path": [],
586
- "model_kwargs": {},
587
- "load_args": false,
588
- "load_data_args": false,
589
- "packing": false,
590
- "packing_length": null,
591
- "packing_num_proc": 1,
592
- "lazy_tokenize": true,
593
- "use_hf": false,
594
- "hub_token": null,
595
- "ddp_timeout": 18000000,
596
- "ddp_backend": null,
597
- "ignore_args_error": false,
598
- "use_swift_lora": false,
599
- "freeze_llm": false,
600
- "freeze_vit": false,
601
- "freeze_aligner": false,
602
- "freeze_parameters": [],
603
- "freeze_parameters_regex": null,
604
- "freeze_parameters_ratio": 0.0,
605
- "trainable_parameters": [],
606
- "trainable_parameters_regex": null,
607
- "adapter_load": null,
608
- "target_modules": [
609
- "all-linear"
610
- ],
611
- "target_regex": null,
612
- "modules_to_save": [],
613
- "lora_rank": 128,
614
- "lora_alpha": 512,
615
- "lora_dropout": 0.05,
616
- "lora_bias": "none",
617
- "lora_dtype": null,
618
- "use_rslora": false,
619
- "rlhf_type": "mapo",
620
- "ref_load": null,
621
- "ref_adapter_load": null,
622
- "beta": 0.001,
623
- "rpo_alpha": null,
624
- "reference_free": false,
625
- "label_smoothing": 0.0,
626
- "f_divergence_type": "reverse_kl",
627
- "desirable_weight": 1.0,
628
- "undesirable_weight": 1.0,
629
- "calculate_KL": null,
630
- "center_rewards_coefficient": null,
631
- "teacher_model": null,
632
- "teacher_model_type": null,
633
- "teacher_model_revision": null,
634
- "lmbda": 0.5,
635
- "seq_kd": false,
636
- "offload_teacher_model": false,
637
- "sft_alpha": 0.0,
638
- "generation_batch_size": 768,
639
- "steps_per_generation": 4,
640
- "num_generations": 8,
641
- "num_generations_eval": null,
642
- "max_completion_length": 1024,
643
- "importance_sampling_level": "token",
644
- "tau_pos": 1.0,
645
- "tau_neg": 1.05,
646
- "epsilon": 0.2,
647
- "epsilon_high": null,
648
- "delta": null,
649
- "use_vllm": true,
650
- "vllm_mode": "server",
651
- "vllm_enable_prefix_caching": true,
652
- "vllm_gpu_memory_utilization": 0.9,
653
- "vllm_tensor_parallel_size": 1,
654
- "vllm_max_model_len": null,
655
- "vllm_enforce_eager": false,
656
- "vllm_limit_mm_per_prompt": null,
657
- "vllm_disable_cascade_attn": false,
658
- "vllm_max_num_seqs": null,
659
- "vllm_mm_processor_cache_gb": null,
660
- "vllm_engine_kwargs": {},
661
- "sleep_level": 0,
662
- "offload_optimizer": false,
663
- "offload_model": true,
664
- "offload_bridge": false,
665
- "vllm_server_base_url": null,
666
- "vllm_server_host": [
667
- "29.164.184.102"
668
- ],
669
- "vllm_server_port": [
670
- 11086
671
- ],
672
- "vllm_server_timeout": 240.0,
673
- "vllm_server_group_port": [
674
- 51313
675
- ],
676
- "reward_funcs": [
677
- "external_mcqa_accuracy",
678
- "external_format",
679
- "external_consistency"
680
- ],
681
- "reward_weights": [
682
- 2.0,
683
- 1.0,
684
- 1.0
685
- ],
686
- "cosine_min_len_value_wrong": -0.5,
687
- "cosine_max_len_value_wrong": 0.0,
688
- "cosine_min_len_value_correct": 1.0,
689
- "cosine_max_len_value_correct": 0.5,
690
- "cosine_max_len": null,
691
- "repetition_n_grams": 3,
692
- "repetition_max_penalty": -1.0,
693
- "soft_max_length": null,
694
- "soft_cache_length": null,
695
- "dynamic_sample": false,
696
- "max_resample_times": 3,
697
- "overlong_filter": true,
698
- "scale_rewards": "group",
699
- "advantage_estimator": "grpo",
700
- "kl_in_reward": false,
701
- "wandb_log_unique_prompts": null,
702
- "log_completions": true,
703
- "rollout_importance_sampling_mode": null,
704
- "rollout_importance_sampling_threshold": 2.0,
705
- "log_rollout_offpolicy_metrics": false,
706
- "off_policy_sequence_mask_delta": null,
707
- "log_entropy": false,
708
- "top_entropy_quantile": 1.0,
709
- "eta": 0.5,
710
- "mapo_advantage_floor_eps": 0.1,
711
- "mapo_task_fail_gate_floor": 0.0,
712
- "mapo_mask_temperature": 1.0,
713
- "mapo_temporal_kappa": 1.0,
714
- "mapo_attention_layers": "43,44,45,46,47",
715
- "mapo_attention_head_reduce": "max",
716
- "mapo_attention_layer_reduce": "mean",
717
- "mapo_failure_reward_name": "MCQAAccuracy",
718
- "mapo_failure_threshold": 0.0,
719
- "mapo_pos_tags": "NOUN,VERB,ADJ,ADV,NUM,X",
720
- "mapo_attention_only": false,
721
- "mapo_debug_attn_grad_probe": false,
722
- "mapo_debug_attn_grad_probe_interval": 0,
723
- "diff_objective": "softplus_margin",
724
- "diff_kl_type": "forward",
725
- "diff_margin_gamma": 2.0,
726
- "diff_margin_beta": 4.0,
727
- "entropy_mask_quantile": 0.25,
728
- "entropy_mask_scope": "sequence",
729
- "entropy_mask_type": "hard",
730
- "text_only_modality_scope": "audio",
731
- "text_ref_model": null,
732
- "text_ref_load": null,
733
- "reward_model": null,
734
- "reward_model_plugin": null,
735
- "sync_ref_model": false,
736
- "ref_model_sync_steps": 512,
737
- "ref_model_mixup_alpha": 0.6,
738
- "async_generate": false,
739
- "move_model_batches": null,
740
- "multi_turn_scheduler": null,
741
- "max_turns": null,
742
- "completion_length_limit_scope": "per_round",
743
- "vllm_server_pass_dataset": false,
744
- "num_iterations": 1,
745
- "loss_type": "grpo",
746
- "check_model": true,
747
- "padded_vocab_size": 152064,
748
- "initialize_embedding": false,
749
- "mlp_padding_free": false,
750
- "load_safetensors": true,
751
- "save_safetensors": true,
752
- "ref_model": null,
753
- "ref_adapters": [],
754
- "merge_lora": false,
755
- "max_shard_size": "5GB",
756
- "train_dataloader_shuffle": true,
757
- "dataloader_pin_memory": true,
758
- "dataloader_persistent_workers": true,
759
- "dataloader_prefetch_factor": 2,
760
- "group_by_length": false,
761
- "llm_model_type": "qwen3_omni_moe_text",
762
- "max_epochs": null,
763
- "enable_dft_loss": false,
764
- "enable_channel_loss": false,
765
- "save_strategy": "steps",
766
- "original_max_position_embeddings": null,
767
- "partial_rotary_factor": null,
768
- "use_shared_expert_gate": false,
769
- "report_to": null,
770
- "vit_gradient_checkpointing": true,
771
- "vit_lr": null,
772
- "aligner_lr": null,
773
- "gradient_checkpointing_kwargs": null,
774
- "linear_num_value_heads": null,
775
- "linear_num_key_heads": null,
776
- "linear_key_head_dim": null,
777
- "linear_value_head_dim": null,
778
- "linear_conv_kernel_dim": null,
779
- "layer_types": null,
780
- "apply_wd_to_qk_layernorm": false,
781
- "mrope_interleaved": true,
782
- "add_version": true
783
- }
784
- }