初始化项目，由ModelHub XC社区提供模型

Model: xd2010/Qwen1.5-MOE-sft-math7k-densemixer
Source: Original Platform

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,61 @@
+---
+base_model: Qwen/Qwen1.5-MoE-A2.7B
+datasets: HectorHe/math7k
+library_name: transformers
+model_name: Qwen1.5-MOE-sft-math7k-densemixer
+tags:
+- generated_from_trainer
+- open-r1
+- densemixer
+- trl
+- sft
+licence: license
+---
+
+# Model Card for Qwen1.5-MOE-sft-math7k-densemixer
+
+This model is a fine-tuned version of [Qwen/Qwen1.5-MoE-A2.7B](https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B) on the [HectorHe/math7k](https://huggingface.co/datasets/HectorHe/math7k) dataset.
+It has been trained using [TRL](https://github.com/huggingface/trl).
+
+## Quick start
+
+```python
+from transformers import pipeline
+
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="xd2010/Qwen1.5-MOE-sft-math7k-densemixer", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
+
+## Training procedure
+
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/hector_-carnegie-mellon-university/huggingface/runs/rwroame6) 
+
+
+This model was trained with SFT.
+
+### Framework versions
+
+- TRL: 0.16.0.dev0
+- Transformers: 4.51.0
+- Pytorch: 2.6.0
+- Datasets: 4.8.4
+- Tokenizers: 0.21.4
+
+## Citations
+
+
+
+Cite TRL as:
+    
+```bibtex
+@misc{vonwerra2022trl,
+	title        = {{TRL: Transformer Reinforcement Learning}},
+	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
+	year         = 2020,
+	journal      = {GitHub repository},
+	publisher    = {GitHub},
+	howpublished = {\url{https://github.com/huggingface/trl}}
+}
+```

added_tokens.json

@@ -0,0 +1,5 @@
+{
+  "<|endoftext|>": 151643,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644
+}

all_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 20312064000.0,
+    "train_loss": 0.8405675888061523,
+    "train_runtime": 27.3869,
+    "train_samples": 6851,
+    "train_samples_per_second": 1.168,
+    "train_steps_per_second": 0.037
+}

config.json

@@ -0,0 +1,38 @@
+{
+  "architectures": [
+    "Qwen2MoeForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "decoder_sparse_step": 1,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5632,
+  "max_position_embeddings": 8192,
+  "max_window_layers": 21,
+  "mlp_only_layers": [],
+  "model_type": "qwen2_moe",
+  "moe_intermediate_size": 1408,
+  "norm_topk_prob": false,
+  "num_attention_heads": 16,
+  "num_experts": 60,
+  "num_experts_per_tok": 4,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 16,
+  "output_router_logits": false,
+  "qkv_bias": true,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "router_aux_loss_coef": 0.001,
+  "shared_expert_intermediate_size": 5632,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,11 @@
+{
+  "attn_implementation": "flash_attention_2",
+  "bos_token_id": 151643,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.51.0",
+  "use_cache": false
+}

model-00001-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6c3f2c964c2527be69f117280eb467cdd24c745aaec1af67bcb49ceacaf6cf1
+size 4996577736

model-00002-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32fa60a8d60ec7e0ac040f2eb62b2256e2fd3e8683fb793314f0076c0d63f079
+size 4996347752

model-00003-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b3bfff5e2f5fbc9d5b90ed1579a2504f22b72a1e15f2ade6ba99c4d7028f396
+size 4997127120

model-00004-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be6eacb56889e40e7aed3ca60bfc45618bcfe5d028c296cf77eb8ce1b111e060
+size 4985592520

model-00005-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6676ad83458ca93f5b23ffee55cea29ed5635fe2b746222c658a0e3727da2366
+size 4996348976

model-00006-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8517fc8ca3d1d46cc2cf005793dbf47131ede0a706b91d93dbc99d24fdd2dfc1
+size 3660151400

special_tokens_map.json

@@ -0,0 +1,14 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<|endoftext|>"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3baadc262572fecb4d3012b01d73309ba5812c5a3cfb69f61c09f8c23aa207d8
+size 11418364

tokenizer_config.json

@@ -0,0 +1,45 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "fast_tokenizer": true,
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 20312064000.0,
+    "train_loss": 0.8405675888061523,
+    "train_runtime": 27.3869,
+    "train_samples": 6851,
+    "train_samples_per_second": 1.168,
+    "train_steps_per_second": 0.037
+}

trainer_state.json

@@ -0,0 +1,51 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.004651162790697674,
+  "eval_steps": 500,
+  "global_step": 1,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.004651162790697674,
+      "grad_norm": 20.563183171334625,
+      "learning_rate": 0.0,
+      "loss": 0.8406,
+      "mean_token_accuracy": 0.824097216129303,
+      "step": 1
+    },
+    {
+      "epoch": 0.004651162790697674,
+      "step": 1,
+      "total_flos": 20312064000.0,
+      "train_loss": 0.8405675888061523,
+      "train_runtime": 27.3869,
+      "train_samples_per_second": 1.168,
+      "train_steps_per_second": 0.037
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 20312064000.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

training.log

@@ -0,0 +1,604 @@
+2026-03-28 22:42:40 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='Qwen/Qwen1.5-MoE-A2.7B', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2026-03-28 22:42:40 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='HectorHe/math7k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2026-03-28 22:42:40 - INFO - __main__ - Training parameters SFTConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=1800000000,
+debug=[],
+deepspeed=None,
+disable_tqdm=False,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=1,
+gradient_checkpointing=True,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Qwen1.5-MOE-sft-math7k-densemixer,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=1e-05,
+length_column_name=length,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer/runs/Mar28_22-42-39_orchard-flame-3,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=4096,
+max_seq_length=None,
+max_steps=1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=1,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=500,
+save_strategy=SaveStrategy.NO,
+save_total_limit=1,
+seed=1234,
+skip_memory_metrics=True,
+system_prompt=None,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tp_size=0,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2026-03-28 22:42:42 - INFO - __main__ - *** Initializing model kwargs ***
+2026-03-28 22:43:28 - INFO - __main__ - ============================================================
+2026-03-28 22:43:28 - INFO - __main__ - DenseMixer: applying MoE forward patch
+2026-03-28 22:43:28 - INFO - __main__ - ============================================================
+2026-03-28 22:43:28 - INFO - __main__ - DenseMixer: patched Qwen2MoeSparseMoeBlock.forward
+2026-03-28 22:43:38 - INFO - __main__ - *** Train ***
+2026-03-28 22:43:38 - INFO - __main__ - Qwen2MoeForCausalLM(
+  (model): Qwen2MoeModel(
+    (embed_tokens): Embedding(151936, 2048)
+    (layers): ModuleList(
+      (0-23): 24 x Qwen2MoeDecoderLayer(
+        (self_attn): Qwen2MoeFlashAttention2(
+          (q_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (k_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (v_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (o_proj): Linear(in_features=2048, out_features=2048, bias=False)
+          (rotary_emb): Qwen2MoeRotaryEmbedding()
+        )
+        (mlp): Qwen2MoeSparseMoeBlock(
+          (gate): Linear(in_features=2048, out_features=60, bias=False)
+          (experts): ModuleList(
+            (0-59): 60 x Qwen2MoeMLP(
+              (gate_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (up_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (down_proj): Linear(in_features=1408, out_features=2048, bias=False)
+              (act_fn): SiLU()
+            )
+          )
+          (shared_expert): Qwen2MoeMLP(
+            (gate_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (up_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (down_proj): Linear(in_features=5632, out_features=2048, bias=False)
+            (act_fn): SiLU()
+          )
+          (shared_expert_gate): Linear(in_features=2048, out_features=1, bias=False)
+        )
+        (input_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+        (post_attention_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+      )
+    )
+    (norm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+    (rotary_emb): Qwen2MoeRotaryEmbedding()
+  )
+  (lm_head): Linear(in_features=2048, out_features=151936, bias=False)
+)
+2026-03-28 22:43:56 - INFO - __main__ - DenseMixer: Using custom dense-backward forward for Qwen2-MoE
+2026-03-28 22:44:21 - INFO - __main__ - *** Save model ***
+2026-03-28 22:46:56 - INFO - __main__ - Model saved to /tmp/data/Qwen1.5-MOE/sft/math7k/densemixer
+2026-03-28 22:46:56 - INFO - __main__ - Pushing to hub...
+2026-03-28 22:49:45 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='Qwen/Qwen1.5-MoE-A2.7B', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2026-03-28 22:49:45 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='HectorHe/math7k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2026-03-28 22:49:45 - INFO - __main__ - Training parameters SFTConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=1800000000,
+debug=[],
+deepspeed=None,
+disable_tqdm=False,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=1,
+gradient_checkpointing=True,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Qwen1.5-MOE-sft-math7k-densemixer,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=1e-05,
+length_column_name=length,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer/runs/Mar28_22-49-45_orchard-flame-3,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=4096,
+max_seq_length=None,
+max_steps=1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=2,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=500,
+save_strategy=SaveStrategy.NO,
+save_total_limit=1,
+seed=1234,
+skip_memory_metrics=True,
+system_prompt=None,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tp_size=0,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2026-03-28 22:49:47 - INFO - __main__ - *** Initializing model kwargs ***
+2026-03-28 22:50:17 - INFO - __main__ - ============================================================
+2026-03-28 22:50:17 - INFO - __main__ - DenseMixer: applying MoE forward patch
+2026-03-28 22:50:17 - INFO - __main__ - ============================================================
+2026-03-28 22:50:17 - INFO - __main__ - DenseMixer: patched Qwen2MoeSparseMoeBlock.forward
+2026-03-28 22:50:19 - INFO - __main__ - *** Train ***
+2026-03-28 22:50:19 - INFO - __main__ - Qwen2MoeForCausalLM(
+  (model): Qwen2MoeModel(
+    (embed_tokens): Embedding(151936, 2048)
+    (layers): ModuleList(
+      (0-23): 24 x Qwen2MoeDecoderLayer(
+        (self_attn): Qwen2MoeFlashAttention2(
+          (q_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (k_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (v_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (o_proj): Linear(in_features=2048, out_features=2048, bias=False)
+          (rotary_emb): Qwen2MoeRotaryEmbedding()
+        )
+        (mlp): Qwen2MoeSparseMoeBlock(
+          (gate): Linear(in_features=2048, out_features=60, bias=False)
+          (experts): ModuleList(
+            (0-59): 60 x Qwen2MoeMLP(
+              (gate_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (up_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (down_proj): Linear(in_features=1408, out_features=2048, bias=False)
+              (act_fn): SiLU()
+            )
+          )
+          (shared_expert): Qwen2MoeMLP(
+            (gate_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (up_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (down_proj): Linear(in_features=5632, out_features=2048, bias=False)
+            (act_fn): SiLU()
+          )
+          (shared_expert_gate): Linear(in_features=2048, out_features=1, bias=False)
+        )
+        (input_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+        (post_attention_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+      )
+    )
+    (norm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+    (rotary_emb): Qwen2MoeRotaryEmbedding()
+  )
+  (lm_head): Linear(in_features=2048, out_features=151936, bias=False)
+)
+2026-03-28 22:50:36 - INFO - __main__ - DenseMixer: Using custom dense-backward forward for Qwen2-MoE
+2026-03-28 22:51:01 - INFO - __main__ - *** Save model ***
+2026-03-28 22:52:11 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='Qwen/Qwen1.5-MoE-A2.7B', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2026-03-28 22:52:11 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='HectorHe/math7k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2026-03-28 22:52:11 - INFO - __main__ - Training parameters SFTConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=1800000000,
+debug=[],
+deepspeed=None,
+disable_tqdm=False,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=1,
+gradient_checkpointing=True,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Qwen1.5-MOE-sft-math7k-densemixer,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=1e-05,
+length_column_name=length,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer/runs/Mar28_22-52-10_orchard-flame-3,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=4096,
+max_seq_length=None,
+max_steps=1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=2,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=/tmp/data/Qwen1.5-MOE/sft/math7k/densemixer,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=500,
+save_strategy=SaveStrategy.NO,
+save_total_limit=1,
+seed=1234,
+skip_memory_metrics=True,
+system_prompt=None,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tp_size=0,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2026-03-28 22:52:12 - INFO - __main__ - *** Initializing model kwargs ***
+2026-03-28 22:52:42 - INFO - __main__ - ============================================================
+2026-03-28 22:52:42 - INFO - __main__ - DenseMixer: applying MoE forward patch
+2026-03-28 22:52:42 - INFO - __main__ - ============================================================
+2026-03-28 22:52:42 - INFO - __main__ - DenseMixer: patched Qwen2MoeSparseMoeBlock.forward
+2026-03-28 22:52:44 - INFO - __main__ - *** Train ***
+2026-03-28 22:52:44 - INFO - __main__ - Qwen2MoeForCausalLM(
+  (model): Qwen2MoeModel(
+    (embed_tokens): Embedding(151936, 2048)
+    (layers): ModuleList(
+      (0-23): 24 x Qwen2MoeDecoderLayer(
+        (self_attn): Qwen2MoeFlashAttention2(
+          (q_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (k_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (v_proj): Linear(in_features=2048, out_features=2048, bias=True)
+          (o_proj): Linear(in_features=2048, out_features=2048, bias=False)
+          (rotary_emb): Qwen2MoeRotaryEmbedding()
+        )
+        (mlp): Qwen2MoeSparseMoeBlock(
+          (gate): Linear(in_features=2048, out_features=60, bias=False)
+          (experts): ModuleList(
+            (0-59): 60 x Qwen2MoeMLP(
+              (gate_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (up_proj): Linear(in_features=2048, out_features=1408, bias=False)
+              (down_proj): Linear(in_features=1408, out_features=2048, bias=False)
+              (act_fn): SiLU()
+            )
+          )
+          (shared_expert): Qwen2MoeMLP(
+            (gate_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (up_proj): Linear(in_features=2048, out_features=5632, bias=False)
+            (down_proj): Linear(in_features=5632, out_features=2048, bias=False)
+            (act_fn): SiLU()
+          )
+          (shared_expert_gate): Linear(in_features=2048, out_features=1, bias=False)
+        )
+        (input_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+        (post_attention_layernorm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+      )
+    )
+    (norm): Qwen2MoeRMSNorm((0,), eps=1e-06)
+    (rotary_emb): Qwen2MoeRotaryEmbedding()
+  )
+  (lm_head): Linear(in_features=2048, out_features=151936, bias=False)
+)
+2026-03-28 22:53:03 - INFO - __main__ - DenseMixer: Using custom dense-backward forward for Qwen2-MoE
+2026-03-28 22:53:28 - INFO - __main__ - *** Save model ***
+2026-03-28 22:56:09 - INFO - __main__ - Model saved to /tmp/data/Qwen1.5-MOE/sft/math7k/densemixer
+2026-03-28 22:56:09 - INFO - __main__ - Pushing to hub...

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a4207ff183b8c492595605710a3f729e5d93aef0487b5e0119a58d5293c02f8
+size 7544