初始化项目，由ModelHub XC社区提供模型

Model: cs-552-2026-flab/safety_model Source: Original Platform
2026-07-07 00:03:32 +08:00
commit 4f39f0af50
19 changed files with 152538 additions and 0 deletions
--- a/hydra/.hydra/config.yaml
+++ b/hydra/.hydra/config.yaml
@@ -0,0 +1,67 @@
+model:
+  hf_id: Qwen/Qwen3-1.7B
+  local_path: /shared-ro/models/Qwen3-1.7B
+  dtype: bfloat16
+  attn_impl: flash_attention_2
+  max_model_len: 4096
+data:
+  name: safety_aug302
+  loader: load_safetybench_distilled
+  kwargs:
+    cache: /scratch/sft_aug302.jsonl
+    limit: null
+training:
+  recipe: sft
+  epochs: 5
+  per_device_train_batch_size: 1
+  gradient_accumulation_steps: 16
+  learning_rate: 5.0e-06
+  warmup_ratio: 0.05
+  lr_scheduler_type: cosine
+  gradient_checkpointing: true
+  bf16: true
+  max_seq_length: 4096
+  logging_steps: 10
+  save_steps: 500
+  save_total_limit: 2
+  report_to: wandb
+  deepspeed: configs/deepspeed_zero2.json
+  use_peft: true
+  lora_r: 16
+  lora_alpha: 32
+  lora_target_modules: q_proj,v_proj,k_proj,o_proj
+  lora_dropout: 0.05
+domain:
+  name: safety
+  thinking_mode: true
+  benchmark: safety
+seed: 42
+run_id: ${now:%Y%m%d_%H%M%S}
+output_dir: /scratch/checkpoints/safety/sft/20260521_aug302
+wandb:
+  project: flab-cs552
+  entity: cs-552-2026-flab
+  mode: null
+  tags: []
+  notes: null
+eval:
+  model_dir: null
+  'n': 8
+  temperature: 0.7
+  top_p: 0.95
+  max_tokens: 4096
+  samples_dir: null
+push:
+  model_dir: null
+es:
+  base_model_dir: null
+  data: null
+hf_push:
+  enabled: false
+  org: cs-552-2026-flab
+  repo: null
+  private: false
+  vllm_validate: true
+  commit_message: null
+validate_after_train: false
+vllm_validate: true
--- a/hydra/.hydra/hydra.yaml
+++ b/hydra/.hydra/hydra.yaml
@@ -0,0 +1,162 @@
+hydra:
+  run:
+    dir: ${output_dir}/hydra
+  sweep:
+    dir: ${output_dir}/hydra_sweep
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - domain=safety
+    - data=safety_aug302
+    - training=sft_distilled_lora
+    - output_dir=/scratch/checkpoints/safety/sft/20260521_aug302
+  job:
+    name: sft
+    chdir: false
+    override_dirname: data=safety_aug302,domain=safety,output_dir=/scratch/checkpoints/safety/sft/20260521_aug302,training=sft_distilled_lora
+    id: ???
+    num: ???
+    config_name: base
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /scratch/team-repo
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /scratch/team-repo/configs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /scratch/checkpoints/safety/sft/20260521_aug302/hydra
+    choices:
+      domain: safety
+      training: sft_distilled_lora
+      data: safety_aug302
+      model: qwen3_17b
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false
--- a/hydra/.hydra/overrides.yaml
+++ b/hydra/.hydra/overrides.yaml
@@ -0,0 +1,4 @@
+- domain=safety
+- data=safety_aug302
+- training=sft_distilled_lora
+- output_dir=/scratch/checkpoints/safety/sft/20260521_aug302
--- a/hydra/sft.log
+++ b/hydra/sft.log
@@ -0,0 +1,189 @@
+[2026-05-21 15:35:13,599][__main__][INFO] - config:
+model:
+  hf_id: Qwen/Qwen3-1.7B
+  local_path: /shared-ro/models/Qwen3-1.7B
+  dtype: bfloat16
+  attn_impl: flash_attention_2
+  max_model_len: 4096
+data:
+  name: safety_aug302
+  loader: load_safetybench_distilled
+  kwargs:
+    cache: /scratch/sft_aug302.jsonl
+    limit: null
+training:
+  recipe: sft
+  epochs: 5
+  per_device_train_batch_size: 1
+  gradient_accumulation_steps: 16
+  learning_rate: 5.0e-06
+  warmup_ratio: 0.05
+  lr_scheduler_type: cosine
+  gradient_checkpointing: true
+  bf16: true
+  max_seq_length: 4096
+  logging_steps: 10
+  save_steps: 500
+  save_total_limit: 2
+  report_to: wandb
+  deepspeed: configs/deepspeed_zero2.json
+  use_peft: true
+  lora_r: 16
+  lora_alpha: 32
+  lora_target_modules: q_proj,v_proj,k_proj,o_proj
+  lora_dropout: 0.05
+domain:
+  name: safety
+  thinking_mode: true
+  benchmark: safety
+seed: 42
+run_id: ${now:%Y%m%d_%H%M%S}
+output_dir: /scratch/checkpoints/safety/sft/20260521_aug302
+wandb:
+  project: flab-cs552
+  entity: cs-552-2026-flab
+  mode: null
+  tags: []
+  notes: null
+eval:
+  model_dir: null
+  'n': 8
+  temperature: 0.7
+  top_p: 0.95
+  max_tokens: 4096
+  samples_dir: null
+push:
+  model_dir: null
+es:
+  base_model_dir: null
+  data: null
+hf_push:
+  enabled: false
+  org: cs-552-2026-flab
+  repo: null
+  private: false
+  vllm_validate: true
+  commit_message: null
+validate_after_train: false
+vllm_validate: true
+
+[2026-05-21 15:35:23,517][flab.shared.repro][WARNING] - wandb.init failed (No API key configured. Use `wandb login` to log in.); continuing without wandb
+[2026-05-21 15:35:23,669][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:35:23,669][huggingface_hub.utils._http][WARNING] - Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+[2026-05-21 15:35:23,677][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:35:23,795][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/tokenizer_config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:35:23,803][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/tokenizer_config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:35:23,917][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B/tree/main/additional_chat_templates?recursive=false&expand=false "HTTP/1.1 404 Not Found"
+[2026-05-21 15:35:24,043][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B/tree/main?recursive=true&expand=false "HTTP/1.1 200 OK"
+[2026-05-21 15:35:24,779][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B "HTTP/1.1 200 OK"
+[2026-05-21 15:35:24,788][__main__][INFO] - Loaded 302 rows from safety.load_safetybench_distilled
+[2026-05-21 15:35:25,165][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:35:25,175][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:35:25,297][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:35:25,309][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:35:25,708][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/generation_config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:35:25,717][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/generation_config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:35:25,757][__main__][WARNING] - bitsandbytes unavailable — falling back to manual layer freezing (equivalent effect: only target projection modules will be trained).
+[2026-05-21 15:35:25,758][__main__][INFO] - Manual freeze: 112 trainable / 198 frozen / 310 total params (target modules: ['k_proj', 'o_proj', 'q_proj', 'v_proj'])
+[2026-05-21 15:35:26,499][__main__][WARNING] - Disabling DeepSpeed: WORLD_SIZE/LOCAL_RANK not set (not launched via torchrun/deepspeed). Use `deepspeed --num_gpus=N` or `torchrun --nproc_per_node=N` to re-enable.
+[2026-05-21 15:35:27,620][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmpt9sryr_o/test.c -o /tmp/tmpt9sryr_o/test.o
+[2026-05-21 15:35:27,642][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmpt9sryr_o/test.o -laio -o /tmp/tmpt9sryr_o/a.out
+[2026-05-21 15:35:27,659][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmpo9kbsup9/test.c -o /tmp/tmpo9kbsup9/test.o
+[2026-05-21 15:35:27,681][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmpo9kbsup9/test.o -L/usr/local/cuda -L/usr/local/cuda/lib64 -lcufile -o /tmp/tmpo9kbsup9/a.out
+[2026-05-21 15:35:27,727][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmppd4b2l4p/test.c -o /tmp/tmppd4b2l4p/test.o
+[2026-05-21 15:35:27,748][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmppd4b2l4p/test.o -laio -o /tmp/tmppd4b2l4p/a.out
+[2026-05-21 15:36:15,765][__main__][INFO] - config:
+model:
+  hf_id: Qwen/Qwen3-1.7B
+  local_path: /shared-ro/models/Qwen3-1.7B
+  dtype: bfloat16
+  attn_impl: flash_attention_2
+  max_model_len: 4096
+data:
+  name: safety_aug302
+  loader: load_safetybench_distilled
+  kwargs:
+    cache: /scratch/sft_aug302.jsonl
+    limit: null
+training:
+  recipe: sft
+  epochs: 5
+  per_device_train_batch_size: 1
+  gradient_accumulation_steps: 16
+  learning_rate: 5.0e-06
+  warmup_ratio: 0.05
+  lr_scheduler_type: cosine
+  gradient_checkpointing: true
+  bf16: true
+  max_seq_length: 4096
+  logging_steps: 10
+  save_steps: 500
+  save_total_limit: 2
+  report_to: wandb
+  deepspeed: configs/deepspeed_zero2.json
+  use_peft: true
+  lora_r: 16
+  lora_alpha: 32
+  lora_target_modules: q_proj,v_proj,k_proj,o_proj
+  lora_dropout: 0.05
+domain:
+  name: safety
+  thinking_mode: true
+  benchmark: safety
+seed: 42
+run_id: ${now:%Y%m%d_%H%M%S}
+output_dir: /scratch/checkpoints/safety/sft/20260521_aug302
+wandb:
+  project: flab-cs552
+  entity: cs-552-2026-flab
+  mode: null
+  tags: []
+  notes: null
+eval:
+  model_dir: null
+  'n': 8
+  temperature: 0.7
+  top_p: 0.95
+  max_tokens: 4096
+  samples_dir: null
+push:
+  model_dir: null
+es:
+  base_model_dir: null
+  data: null
+hf_push:
+  enabled: false
+  org: cs-552-2026-flab
+  repo: null
+  private: false
+  vllm_validate: true
+  commit_message: null
+validate_after_train: false
+vllm_validate: true
+
+[2026-05-21 15:36:26,944][flab.shared.repro][INFO] - wandb run: safety__sft__20260521_153621 (https://wandb.ai/cs-552-2026-flab/flab-cs552/runs/893zqwbg)
+[2026-05-21 15:36:27,097][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:36:27,105][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:36:27,222][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/tokenizer_config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:36:27,222][huggingface_hub.utils._http][WARNING] - Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+[2026-05-21 15:36:27,231][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/tokenizer_config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:36:27,349][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B/tree/main/additional_chat_templates?recursive=false&expand=false "HTTP/1.1 404 Not Found"
+[2026-05-21 15:36:27,463][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B/tree/main?recursive=true&expand=false "HTTP/1.1 200 OK"
+[2026-05-21 15:36:28,184][httpx][INFO] - HTTP Request: GET https://huggingface.co/api/models/Qwen/Qwen3-1.7B "HTTP/1.1 200 OK"
+[2026-05-21 15:36:28,194][__main__][INFO] - Loaded 302 rows from safety.load_safetybench_distilled
+[2026-05-21 15:36:28,559][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:36:28,572][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:36:28,690][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:36:28,696][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:36:29,090][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/Qwen/Qwen3-1.7B/resolve/main/generation_config.json "HTTP/1.1 307 Temporary Redirect"
+[2026-05-21 15:36:29,099][httpx][INFO] - HTTP Request: HEAD https://huggingface.co/api/resolve-cache/models/Qwen/Qwen3-1.7B/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/generation_config.json "HTTP/1.1 200 OK"
+[2026-05-21 15:36:29,143][__main__][WARNING] - bitsandbytes unavailable — falling back to manual layer freezing (equivalent effect: only target projection modules will be trained).
+[2026-05-21 15:36:29,144][__main__][INFO] - Manual freeze: 112 trainable / 198 frozen / 310 total params (target modules: ['k_proj', 'o_proj', 'q_proj', 'v_proj'])
+[2026-05-21 15:36:29,898][__main__][WARNING] - Disabling DeepSpeed: WORLD_SIZE/LOCAL_RANK not set (not launched via torchrun/deepspeed). Use `deepspeed --num_gpus=N` or `torchrun --nproc_per_node=N` to re-enable.
+[2026-05-21 15:36:31,057][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmp9nakbki7/test.c -o /tmp/tmp9nakbki7/test.o
+[2026-05-21 15:36:31,079][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmp9nakbki7/test.o -laio -o /tmp/tmp9nakbki7/a.out
+[2026-05-21 15:36:31,096][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmpcejqunra/test.c -o /tmp/tmpcejqunra/test.o
+[2026-05-21 15:36:31,117][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmpcejqunra/test.o -L/usr/local/cuda -L/usr/local/cuda/lib64 -lcufile -o /tmp/tmpcejqunra/a.out
+[2026-05-21 15:36:31,166][root][INFO] - x86_64-linux-gnu-gcc -fno-strict-overflow -Wsign-compare -DNDEBUG -g -O2 -Wall -fPIC -c /tmp/tmp3shf_f2c/test.c -o /tmp/tmp3shf_f2c/test.o
+[2026-05-21 15:36:31,187][root][INFO] - x86_64-linux-gnu-gcc /tmp/tmp3shf_f2c/test.o -laio -o /tmp/tmp3shf_f2c/a.out
+[2026-05-21 15:41:47,398][__main__][INFO] - Saved checkpoint to /scratch/checkpoints/safety/sft/20260521_aug302