Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,12 @@ saved_model/**/* 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
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.0-of-4.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.0-of-4.jsonl filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.1-of-4.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.1-of-4.jsonl filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.2-of-4.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.2-of-4.jsonl filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.3-of-4.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.3-of-4.jsonl filter=lfs diff=lfs merge=lfs -text
+checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.decomposed.dpo.json filter=lfs diff=lfs merge=lfs -text

checkpoint-100/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 32,
+  "pad_token_id": 100001,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": false,
+  "vocab_size": 102400
+}

checkpoint-100/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "pad_token_id": 100001,
+  "transformers_version": "4.40.1"
+}

checkpoint-100/math500.test.wrong_correct.v1.0.0shot.n5.tem1.0.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.10869565217391304,
+  "pass@k": 0.32608695652173914,
+  "correct": 30,
+  "total": 276
+}

checkpoint-100/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4fc4fd23ef6e7b66a83899b97362d5132b53e85184416c98e7d335eec19f0e7
+size 13820769582

checkpoint-100/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-100/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "100000": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100001": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100002": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{% if message['role'] == 'user' %}{{ 'User: ' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Assistant: ' + message['content'] + eos_token }}{% elif message['role'] == 'system' %}{{ message['content'] + '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "padding_side": "left",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-100/training_config.yaml

@@ -0,0 +1,144 @@
+ds_cfg:
+  train_micro_batch_size_per_gpu: ${per_gpu_train_batch_size}
+  gradient_accumulation_steps: ${gradient_accumulation_steps}
+  scheduler:
+    type: WarmupDecayLR
+    params:
+      total_num_steps: 404
+      warmup_max_lr: ${learning_rate}
+      warmup_num_steps: 24
+      warmup_type: linear
+  optimizer:
+    type: AdamW
+    params:
+      lr: ${learning_rate}
+      betas:
+      - 0.9
+      - 0.95
+      eps: 1.0e-06
+      weight_decay: ${weight_decay}
+  bf16:
+    enabled: true
+  zero_optimization:
+    stage: 1
+    offload_optimizer:
+      device: cpu
+      pin_memory: true
+    stage3_param_persistence_threshold: 100000.0
+    stage3_max_live_parameters: 100000000.0
+    stage3_prefetch_bucket_size: 100000000.0
+    memory_efficient_linear: false
+  steps_per_print: 25
+  gradient_clipping: 1.0
+  prescale_gradients: false
+sft_model_dir: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+train_file: ${sft_model_dir}/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.decomposed.json
+dev_file: null
+test_file: null
+torch_dtype:
+  _target_: general_util.training_utils.return_torch_dtype
+  dtype: bfloat16
+tokenizer_init:
+  _target_: general_util.tokenization_utils.init_tokenizer
+  tokenizer_path: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+  padding_side: left
+device_map:
+  _target_: models.utils.return_single_device_map
+model:
+  _target_: models.llama.LlamaForCausalLMDPO.from_pretrained_with_ref_model
+  beta: 0.5
+  gradient_checkpointing: true
+  attn_implementation: flash_attention_2
+  torch_dtype: ${torch_dtype}
+  device_map: ${device_map}
+  ref_model:
+    _target_: models.llama.LlamaForCausalLMDPO.from_pretrained
+    pretrained_model_name_or_path: ${model_name_or_path}
+    torch_dtype: ${torch_dtype}
+    attn_implementation: flash_attention_2
+    device_map: ${device_map}
+read_tensor:
+  _target_: data.logic_combine.MultiMappingDataset
+  aligner:
+    _target_: data.input_aligner.concat_aligner
+    aligners:
+    - _target_: data.input_aligner.dpo_pair_aligner
+      pos_field: chosen_response
+      neg_field: reject_response
+  template:
+    chosen: '{text}{pos}<｜end▁of▁sentence｜>'
+    reject: '{text}{neg}<｜end▁of▁sentence｜>'
+    prompt: '{text}'
+  kv_mapping:
+    chosen: chosen
+    reject: reject
+    id: index
+    prompt: prompt
+dist_load_data_barrier: false
+extended_vocab: null
+collator:
+  _target_: data.dpo.DPOCollator
+  tokenizer: ${tokenizer_init}
+  max_seq_length: 1024
+num_workers: 8
+prefetch_factor: 2
+model_name_or_path: ${sft_model_dir}
+pretrain: null
+dp_size: 4
+tp_size: 1
+pp_size: 1
+exp_name: deepseek-math.7b.ins.meta_math_cot.math55k.n5.critic_correct.dpo.H100.w4.v2.0.s${seed}
+exp_notes: null
+output_dir: experiments/${exp_name}
+do_train: true
+evaluate_during_training: false
+do_eval: false
+eval_sub_path: checkpoint-100
+per_gpu_train_batch_size: 4
+per_gpu_eval_batch_size: 4
+learning_rate: 1.0e-06
+gradient_accumulation_steps: 4
+weight_decay: 0.1
+adam_epsilon: 1.0e-06
+adam_betas: (0.9, 0.98)
+total_dataset_len: 25864
+max_grad_norm: 1.0
+num_train_epochs: 1
+max_steps: 0
+warmup_proportion: 0.06
+warmup_steps: 0
+optimizer: null
+use_nvlamb: null
+bit_training: null
+logging_steps: 5
+save_ds_state: false
+save_steps: 100
+save_best: false
+eval_steps: 400
+ddp_eval: true
+no_cuda: false
+seed: 42
+local_rank: 0
+fp16: true
+fp16_opt_level: O1
+fp16_bfloat16: true
+prediction_cfg:
+  metric: loss
+  measure: -1
+  best_checkpoint: null
+  best_result: null
+eval_forward_fn:
+  _target_: general_util.evaluator.DefaultForwardFn
+post_process:
+  _target_: post_processors.dpo.DPOEvalPostProcessor
+summary_helper:
+  _target_: general_util.tensorboard_helper.WandbWriter
+  batch_index_or_keys: null
+  outputs_index_or_keys:
+    train/chosen_reward: chosen_reward
+    train/rejected_reward: rejected_reward
+n_gpu: 1
+device: cuda:0
+train_batch_size: 4
+eval_batch_size: null
+world_size: 4

checkpoint-200/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 32,
+  "pad_token_id": 100001,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": false,
+  "vocab_size": 102400
+}

checkpoint-200/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "pad_token_id": 100001,
+  "transformers_version": "4.40.1"
+}

checkpoint-200/math500.test.wrong_correct.v1.0.0shot.n5.tem1.0.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.10144927536231885,
+  "pass@k": 0.2898550724637681,
+  "correct": 28,
+  "total": 276
+}

checkpoint-200/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99a1e19619c6c86bc6ca9d86c031d1ab5dce94f9229e1227d65992f88c2217a3
+size 13820769582

checkpoint-200/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-200/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "100000": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100001": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100002": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{% if message['role'] == 'user' %}{{ 'User: ' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Assistant: ' + message['content'] + eos_token }}{% elif message['role'] == 'system' %}{{ message['content'] + '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "padding_side": "left",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-200/training_config.yaml

@@ -0,0 +1,144 @@
+ds_cfg:
+  train_micro_batch_size_per_gpu: ${per_gpu_train_batch_size}
+  gradient_accumulation_steps: ${gradient_accumulation_steps}
+  scheduler:
+    type: WarmupDecayLR
+    params:
+      total_num_steps: 404
+      warmup_max_lr: ${learning_rate}
+      warmup_num_steps: 24
+      warmup_type: linear
+  optimizer:
+    type: AdamW
+    params:
+      lr: ${learning_rate}
+      betas:
+      - 0.9
+      - 0.95
+      eps: 1.0e-06
+      weight_decay: ${weight_decay}
+  bf16:
+    enabled: true
+  zero_optimization:
+    stage: 1
+    offload_optimizer:
+      device: cpu
+      pin_memory: true
+    stage3_param_persistence_threshold: 100000.0
+    stage3_max_live_parameters: 100000000.0
+    stage3_prefetch_bucket_size: 100000000.0
+    memory_efficient_linear: false
+  steps_per_print: 25
+  gradient_clipping: 1.0
+  prescale_gradients: false
+sft_model_dir: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+train_file: ${sft_model_dir}/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.decomposed.json
+dev_file: null
+test_file: null
+torch_dtype:
+  _target_: general_util.training_utils.return_torch_dtype
+  dtype: bfloat16
+tokenizer_init:
+  _target_: general_util.tokenization_utils.init_tokenizer
+  tokenizer_path: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+  padding_side: left
+device_map:
+  _target_: models.utils.return_single_device_map
+model:
+  _target_: models.llama.LlamaForCausalLMDPO.from_pretrained_with_ref_model
+  beta: 0.5
+  gradient_checkpointing: true
+  attn_implementation: flash_attention_2
+  torch_dtype: ${torch_dtype}
+  device_map: ${device_map}
+  ref_model:
+    _target_: models.llama.LlamaForCausalLMDPO.from_pretrained
+    pretrained_model_name_or_path: ${model_name_or_path}
+    torch_dtype: ${torch_dtype}
+    attn_implementation: flash_attention_2
+    device_map: ${device_map}
+read_tensor:
+  _target_: data.logic_combine.MultiMappingDataset
+  aligner:
+    _target_: data.input_aligner.concat_aligner
+    aligners:
+    - _target_: data.input_aligner.dpo_pair_aligner
+      pos_field: chosen_response
+      neg_field: reject_response
+  template:
+    chosen: '{text}{pos}<｜end▁of▁sentence｜>'
+    reject: '{text}{neg}<｜end▁of▁sentence｜>'
+    prompt: '{text}'
+  kv_mapping:
+    chosen: chosen
+    reject: reject
+    id: index
+    prompt: prompt
+dist_load_data_barrier: false
+extended_vocab: null
+collator:
+  _target_: data.dpo.DPOCollator
+  tokenizer: ${tokenizer_init}
+  max_seq_length: 1024
+num_workers: 8
+prefetch_factor: 2
+model_name_or_path: ${sft_model_dir}
+pretrain: null
+dp_size: 4
+tp_size: 1
+pp_size: 1
+exp_name: deepseek-math.7b.ins.meta_math_cot.math55k.n5.critic_correct.dpo.H100.w4.v2.0.s${seed}
+exp_notes: null
+output_dir: experiments/${exp_name}
+do_train: true
+evaluate_during_training: false
+do_eval: false
+eval_sub_path: checkpoint-100
+per_gpu_train_batch_size: 4
+per_gpu_eval_batch_size: 4
+learning_rate: 1.0e-06
+gradient_accumulation_steps: 4
+weight_decay: 0.1
+adam_epsilon: 1.0e-06
+adam_betas: (0.9, 0.98)
+total_dataset_len: 25864
+max_grad_norm: 1.0
+num_train_epochs: 1
+max_steps: 0
+warmup_proportion: 0.06
+warmup_steps: 0
+optimizer: null
+use_nvlamb: null
+bit_training: null
+logging_steps: 5
+save_ds_state: false
+save_steps: 100
+save_best: false
+eval_steps: 400
+ddp_eval: true
+no_cuda: false
+seed: 42
+local_rank: 0
+fp16: true
+fp16_opt_level: O1
+fp16_bfloat16: true
+prediction_cfg:
+  metric: loss
+  measure: -1
+  best_checkpoint: null
+  best_result: null
+eval_forward_fn:
+  _target_: general_util.evaluator.DefaultForwardFn
+post_process:
+  _target_: post_processors.dpo.DPOEvalPostProcessor
+summary_helper:
+  _target_: general_util.tensorboard_helper.WandbWriter
+  batch_index_or_keys: null
+  outputs_index_or_keys:
+    train/chosen_reward: chosen_reward
+    train/rejected_reward: rejected_reward
+n_gpu: 1
+device: cuda:0
+train_batch_size: 4
+eval_batch_size: null
+world_size: 4

checkpoint-300/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 32,
+  "pad_token_id": 100001,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": false,
+  "vocab_size": 102400
+}

checkpoint-300/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "pad_token_id": 100001,
+  "transformers_version": "4.40.1"
+}

checkpoint-300/math500.test.wrong_correct.v1.0.0shot.n5.tem1.0.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.14130434782608695,
+  "pass@k": 0.3115942028985507,
+  "correct": 39,
+  "total": 276
+}

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.0-of-4.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0f3b7706cdd3548a4654f8b2dda9f4e0969cd3ba11887e48d2db359219f2ed3
+size 24658738

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.0-of-4.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49bae6b021243e6c8293a5c9f8613a6ad3e9ee6a9f46cd59825bc4fb895ffcfe
+size 24261798

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.0-of-4.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.3133608815426997,
+  "pass@k": 0.5650826446280992,
+  "correct": 910,
+  "total": 2904
+}

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.1-of-4.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f771fbf612c6827707a8377ed614d316e43e360d1c29abffacc3281ec2843979
+size 25073186

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.1-of-4.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9655dfa50808353edfd8956044acaf85ab0a18849131513aa8cfc450fe3fecbe
+size 24676309

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.1-of-4.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.33505509641873277,
+  "pass@k": 0.5702479338842975,
+  "correct": 973,
+  "total": 2904
+}

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.2-of-4.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b697655ea6a42ec440866406701502d81a02d49a46473d5bae69f2b1d83ec34
+size 25072252

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.2-of-4.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fffb79729b9d5c108904d900326760713c87ad7042c822d9bbb8e8522be948b9
+size 24675394

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.2-of-4.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.3415977961432507,
+  "pass@k": 0.581267217630854,
+  "correct": 992,
+  "total": 2904
+}

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.3-of-4.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87a0b5b90a2152dfbb524cdb5af0e48e7e8100ca11a37d06e27768ecc55e8a7a
+size 24848726

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.3-of-4.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ff23a705afc98563b8b64354c06df79493628c8c5fb45439db9ca83f3dd7cf1
+size 24452130

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.3-of-4.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.33769813921433495,
+  "pass@k": 0.5678842177808407,
+  "correct": 980,
+  "total": 2902
+}

checkpoint-300/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.decomposed.dpo.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00c2c8efe212cb48172dbabec60545eb301ea7cd0dde45a50e2e786830f0d8ad
+size 264188903

checkpoint-300/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7238d3011e15d2c96e43e1b3bca3014cc87788eb8be5fb27b22cdc0a8a7274ba
+size 13820769582

checkpoint-300/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-300/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "100000": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100001": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100002": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{{ bos_token }}{% for message in messages %}{% if message['role'] == 'user' %}{{ 'User: ' + message['content'] + '\n\n' }}{% elif message['role'] == 'assistant' %}{{ 'Assistant: ' + message['content'] + eos_token }}{% elif message['role'] == 'system' %}{{ message['content'] + '\n\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "padding_side": "left",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-300/training_config.yaml

@@ -0,0 +1,144 @@
+ds_cfg:
+  train_micro_batch_size_per_gpu: ${per_gpu_train_batch_size}
+  gradient_accumulation_steps: ${gradient_accumulation_steps}
+  scheduler:
+    type: WarmupDecayLR
+    params:
+      total_num_steps: 404
+      warmup_max_lr: ${learning_rate}
+      warmup_num_steps: 24
+      warmup_type: linear
+  optimizer:
+    type: AdamW
+    params:
+      lr: ${learning_rate}
+      betas:
+      - 0.9
+      - 0.95
+      eps: 1.0e-06
+      weight_decay: ${weight_decay}
+  bf16:
+    enabled: true
+  zero_optimization:
+    stage: 1
+    offload_optimizer:
+      device: cpu
+      pin_memory: true
+    stage3_param_persistence_threshold: 100000.0
+    stage3_max_live_parameters: 100000000.0
+    stage3_prefetch_bucket_size: 100000000.0
+    memory_efficient_linear: false
+  steps_per_print: 25
+  gradient_clipping: 1.0
+  prescale_gradients: false
+sft_model_dir: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+train_file: ${sft_model_dir}/meta_math_sub_math.55k.cot.n5.tem0.8.p0.9.v1.0_dsk_clean_v2.wrong_correct.v1.0.0shot.n5.tem1.0.decomposed.json
+dev_file: null
+test_file: null
+torch_dtype:
+  _target_: general_util.training_utils.return_torch_dtype
+  dtype: bfloat16
+tokenizer_init:
+  _target_: general_util.tokenization_utils.init_tokenizer
+  tokenizer_path: experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200
+  padding_side: left
+device_map:
+  _target_: models.utils.return_single_device_map
+model:
+  _target_: models.llama.LlamaForCausalLMDPO.from_pretrained_with_ref_model
+  beta: 0.5
+  gradient_checkpointing: true
+  attn_implementation: flash_attention_2
+  torch_dtype: ${torch_dtype}
+  device_map: ${device_map}
+  ref_model:
+    _target_: models.llama.LlamaForCausalLMDPO.from_pretrained
+    pretrained_model_name_or_path: ${model_name_or_path}
+    torch_dtype: ${torch_dtype}
+    attn_implementation: flash_attention_2
+    device_map: ${device_map}
+read_tensor:
+  _target_: data.logic_combine.MultiMappingDataset
+  aligner:
+    _target_: data.input_aligner.concat_aligner
+    aligners:
+    - _target_: data.input_aligner.dpo_pair_aligner
+      pos_field: chosen_response
+      neg_field: reject_response
+  template:
+    chosen: '{text}{pos}<｜end▁of▁sentence｜>'
+    reject: '{text}{neg}<｜end▁of▁sentence｜>'
+    prompt: '{text}'
+  kv_mapping:
+    chosen: chosen
+    reject: reject
+    id: index
+    prompt: prompt
+dist_load_data_barrier: false
+extended_vocab: null
+collator:
+  _target_: data.dpo.DPOCollator
+  tokenizer: ${tokenizer_init}
+  max_seq_length: 1024
+num_workers: 8
+prefetch_factor: 2
+model_name_or_path: ${sft_model_dir}
+pretrain: null
+dp_size: 4
+tp_size: 1
+pp_size: 1
+exp_name: deepseek-math.7b.ins.meta_math_cot.math55k.n5.critic_correct.dpo.H100.w4.v2.0.s${seed}
+exp_notes: null
+output_dir: experiments/${exp_name}
+do_train: true
+evaluate_during_training: false
+do_eval: false
+eval_sub_path: checkpoint-100
+per_gpu_train_batch_size: 4
+per_gpu_eval_batch_size: 4
+learning_rate: 1.0e-06
+gradient_accumulation_steps: 4
+weight_decay: 0.1
+adam_epsilon: 1.0e-06
+adam_betas: (0.9, 0.98)
+total_dataset_len: 25864
+max_grad_norm: 1.0
+num_train_epochs: 1
+max_steps: 0
+warmup_proportion: 0.06
+warmup_steps: 0
+optimizer: null
+use_nvlamb: null
+bit_training: null
+logging_steps: 5
+save_ds_state: false
+save_steps: 100
+save_best: false
+eval_steps: 400
+ddp_eval: true
+no_cuda: false
+seed: 42
+local_rank: 0
+fp16: true
+fp16_opt_level: O1
+fp16_bfloat16: true
+prediction_cfg:
+  metric: loss
+  measure: -1
+  best_checkpoint: null
+  best_result: null
+eval_forward_fn:
+  _target_: general_util.evaluator.DefaultForwardFn
+post_process:
+  _target_: post_processors.dpo.DPOEvalPostProcessor
+summary_helper:
+  _target_: general_util.tensorboard_helper.WandbWriter
+  batch_index_or_keys: null
+  outputs_index_or_keys:
+    train/chosen_reward: chosen_reward
+    train/rejected_reward: rejected_reward
+n_gpu: 1
+device: cuda:0
+train_batch_size: 4
+eval_batch_size: null
+world_size: 4

checkpoint-400/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "experiments/deepseek-math.ins.7b.assess.sft.H100.w2.v2.2/checkpoint-200",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 32,
+  "pad_token_id": 100001,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.40.1",
+  "use_cache": false,
+  "vocab_size": 102400
+}

checkpoint-400/generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "pad_token_id": 100001,
+  "transformers_version": "4.40.1"
+}

checkpoint-400/math500.test.wrong_correct.v1.0.0shot.n5.tem1.0.metrics.json

@@ -0,0 +1,6 @@
+{
+  "acc": 0.12681159420289856,
+  "pass@k": 0.2898550724637681,
+  "correct": 35,
+  "total": 276
+}