Upload folder using huggingface_hub

.gitattributes

@@ -59,3 +59,13 @@ grpo_qwen_14b/best_adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 grpo_qwen_14b/checkpoints/checkpoint-400/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 grpo_qwen_14b/checkpoints/checkpoint-500/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 grpo_qwen_14b/wandb/run-20251227_194423-jz7bptqa/run-jz7bptqa.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/run-fb8js9es.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_153517-g5bybskm/run-g5bybskm.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_153729-pmpxe28f/run-pmpxe28f.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_153949-6fxdx0d2/run-6fxdx0d2.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_154144-0ek9e5bk/run-0ek9e5bk.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_154334-wvpf8qeo/run-wvpf8qeo.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_154526-q26c0nv5/run-q26c0nv5.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_154823-csl0hdpv/run-csl0hdpv.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_155025-xzbi1gai/run-xzbi1gai.wandb filter=lfs diff=lfs merge=lfs -text
+dpo_run_24b_v1/wandb/run-20251226_155204-00msx40b/run-00msx40b.wandb filter=lfs diff=lfs merge=lfs -text

dpo_run_24b_v1/config_resolved.yaml

@@ -0,0 +1,93 @@
+run:
+  run_dir: ./runs/dpo_run_24b_v1
+  seed: 42
+wandb:
+  enabled: true
+  project: dpo-training
+  entity: null
+  name: null
+  tags:
+  - dpo-lora
+  - preference-optimization
+  notes: null
+model:
+  repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+  revision: null
+  base_local_dir: base_model
+  trust_remote_code: true
+  tokenizer_use_fast: true
+  device_map: auto
+  torch_dtype: bfloat16
+  use_4bit: false
+  bnb_4bit_quant_type: nf4
+  bnb_4bit_use_double_quant: false
+  bnb_4bit_compute_dtype: bfloat16
+  attn_implementation: null
+data:
+  train_jsonl: dpo_pairs_generated.jsonl
+  eval_jsonl: null
+  eval_split_ratio: 0.1
+  prompt_field: prompt
+  chosen_field: chosen
+  rejected_field: rejected
+  score_field: f1_score
+  format_type: chatml
+  system_prompt: "You are a Hyperswitch Rust code analyzer. Identify functions/structs\
+    \ that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain\
+    \ the data flow and why each component must change:\n- Flow: [Input \u2192 Processing\
+    \ \u2192 Output with arrows]\n- For each component: \"The [ComponentName] ([path])\
+    \ must [action] because [reason]\u2014without this, [consequence]\"\n- Explain\
+    \ coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\n\
+    add::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n\
+    1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for\
+    \ nested items: `status::StructName::Type::Name`\n3. Always explain \"must change\
+    \ because\" and \"without this\"\n3. Types of components: function, struct, enum,\
+    \ impl, trait\n4. If there is extra information (e.g., enum variants), include\
+    \ that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n"
+  max_length: 2048
+  shuffle: true
+  num_proc: 4
+peft:
+  enabled: true
+  r: 16
+  lora_alpha: 32
+  lora_dropout: 0.05
+  bias: none
+  target_modules: auto
+dpo:
+  beta: 0.1
+  label_smoothing: 0.0
+  loss_type: sigmoid
+  use_reference_model: true
+  reference_free: false
+train:
+  num_train_epochs: 3
+  per_device_train_batch_size: 1
+  per_device_eval_batch_size: 1
+  gradient_accumulation_steps: 8
+  learning_rate: 5e-5
+  weight_decay: 0.0
+  warmup_ratio: 0.1
+  lr_scheduler_type: cosine
+  optim: adamw_torch
+  max_grad_norm: 1.0
+  gradient_checkpointing: true
+  logging_steps: 2
+  save_strategy: steps
+  save_steps: 100
+  save_total_limit: 10
+  evaluation_strategy: steps
+  eval_steps: 25
+  load_best_model_at_end: true
+  early_stopping:
+    enabled: true
+    patience: 5
+    min_delta: 0.001
+    metric: eval_loss
+    mode: min
+  resume_from_checkpoint: auto
+merge:
+  enabled: true
+  merged_dtype: float16
+  max_shard_size: 2GB
+  output_dir: ./merged_14b_dpo_lora

dpo_run_24b_v1/wandb/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:52:04.520208686Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:52:04.677441728Z","level":"INFO","msg":"stream: created new stream","id":"00msx40b"}
+{"time":"2025-12-26T15:52:04.677538624Z","level":"INFO","msg":"handler: started","stream_id":"00msx40b"}
+{"time":"2025-12-26T15:52:04.677662488Z","level":"INFO","msg":"stream: started","id":"00msx40b"}
+{"time":"2025-12-26T15:52:04.677684998Z","level":"INFO","msg":"writer: started","stream_id":"00msx40b"}
+{"time":"2025-12-26T15:52:04.677696651Z","level":"INFO","msg":"sender: started","stream_id":"00msx40b"}
+{"time":"2025-12-26T15:52:45.509029743Z","level":"INFO","msg":"stream: closing","id":"00msx40b"}
+{"time":"2025-12-26T15:52:45.704898985Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:52:45.822151941Z","level":"INFO","msg":"handler: closed","stream_id":"00msx40b"}
+{"time":"2025-12-26T15:52:45.822254749Z","level":"INFO","msg":"sender: closed","stream_id":"00msx40b"}
+{"time":"2025-12-26T15:52:45.822266001Z","level":"INFO","msg":"stream: closed","id":"00msx40b"}

dpo_run_24b_v1/wandb/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_setup.py:_flush():80] Configure stats pid to 147336
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_155204-00msx40b/logs/debug.log
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_155204-00msx40b/logs/debug-internal.log
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:52:04,246 INFO    MainThread:147336 [wandb_init.py:init():889] starting backend
+2025-12-26 15:52:04,513 INFO    MainThread:147336 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:52:04,518 INFO    MainThread:147336 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:52:04,520 INFO    MainThread:147336 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:52:04,521 INFO    MainThread:147336 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:52:04,863 INFO    MainThread:147336 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:52:04,981 INFO    MainThread:147336 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:52:04,981 INFO    MainThread:147336 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:52:04,981 INFO    MainThread:147336 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:52:04,981 INFO    MainThread:147336 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:52:04,987 INFO    MainThread:147336 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:52:45,509 INFO    wandb-AsyncioManager-main:147336 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:52:45,509 INFO    wandb-AsyncioManager-main:147336 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            05zaxpswkd8t9j7nmlszf02dmdnkyst5:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "315579105280"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:26:29.019842Z"
+                writerId: 05zaxpswkd8t9j7nmlszf02dmdnkyst5
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/files/output.log

@@ -0,0 +1,44 @@
+Wandb initialized: project='dpo-training', name='auto-generated'
+`torch_dtype` is deprecated! Use `dtype` instead!
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 555, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 953, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 744, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 568, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 555, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 953, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 744, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 568, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/files/requirements.txt

@@ -0,0 +1,104 @@
+exceptiongroup==1.3.1
+wheel==0.45.1
+python-dateutil==2.9.0.post0
+nvidia-ml-py==13.580.82
+huggingface_hub==1.2.3
+idna==3.11
+click==8.3.1
+numpy==2.2.6
+httpx==0.28.1
+tokenizers==0.22.1
+sympy==1.13.1
+yarl==1.22.0
+async-timeout==5.0.1
+datasets==4.4.2
+platformdirs==4.5.1
+nvidia-cuda-cupti-cu12==12.1.105
+nvidia-nvtx-cu12==12.1.105
+smmap==5.0.2
+accelerate==1.12.0
+requests==2.32.5
+aiohttp==3.13.2
+bitsandbytes==0.49.0
+nvidia-cublas-cu12==12.1.3.1
+mpmath==1.3.0
+torchaudio==2.5.1+cu121
+nvidia-cuda-runtime-cu12==12.1.105
+typing-inspection==0.4.2
+GitPython==3.1.45
+xxhash==3.6.0
+nvidia-cusolver-cu12==11.4.5.107
+pydantic_core==2.41.5
+six==1.17.0
+torchvision==0.20.1+cu121
+typing_extensions==4.15.0
+triton==3.1.0
+charset-normalizer==3.4.4
+nvitop==1.6.1
+wandb==0.23.1
+regex==2025.11.3
+pip==25.3
+nvidia-cusparse-cu12==12.1.0.106
+pytz==2025.2
+Jinja2==3.1.6
+psutil==7.2.0
+pillow==12.0.0
+packaging==25.0
+safetensors==0.7.0
+sentry-sdk==2.48.0
+gitdb==4.0.12
+httpcore==1.0.9
+setuptools==80.9.0
+nvidia-cufft-cu12==11.0.2.54
+anyio==4.12.0
+transformers==5.0.0.dev0
+pydantic==2.12.5
+fsspec==2025.10.0
+filelock==3.20.0
+PyYAML==6.0.3
+hf-xet==1.2.0
+nvidia-cudnn-cu12==9.1.0.70
+tqdm==4.67.1
+MarkupSafe==2.1.5
+attrs==25.4.0
+nvidia-cuda-nvrtc-cu12==12.1.105
+peft==0.18.0
+aiohappyeyeballs==2.6.1
+networkx==3.4.2
+nvidia-nvjitlink-cu12==12.9.86
+certifi==2025.11.12
+pyarrow==22.0.0
+dill==0.4.0
+protobuf==6.33.2
+aiosignal==1.4.0
+frozenlist==1.8.0
+urllib3==2.6.2
+propcache==0.4.1
+tzdata==2025.3
+pandas==2.3.3
+annotated-types==0.7.0
+shellingham==1.5.4
+nvidia-nccl-cu12==2.21.5
+multidict==6.7.0
+nvidia-curand-cu12==10.3.2.106
+trl==0.26.2
+torch==2.5.1+cu121
+h11==0.16.0
+multiprocess==0.70.18
+typer-slim==0.21.0
+wheel==0.45.1
+tomli==2.0.1
+autocommand==2.2.2
+jaraco.context==5.3.0
+zipp==3.19.2
+packaging==24.2
+inflect==7.3.1
+typing_extensions==4.12.2
+platformdirs==4.2.2
+jaraco.functools==4.0.1
+jaraco.collections==5.1.0
+jaraco.text==3.12.1
+backports.tarfile==1.2.0
+more-itertools==10.3.0
+importlib_metadata==8.0.0
+typeguard==4.3.0

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/files/wandb-metadata.json

@@ -0,0 +1,47 @@
+{
+  "os":  "Linux-6.12.46+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.10.12",
+  "startedAt":  "2025-12-26T15:26:29.019842Z",
+  "args":  [
+    "--config",
+    "config_dpo.yaml"
+  ],
+  "program":  "/workspace/trainer-kit/DPO/run_dpo.py",
+  "codePath":  "run_dpo.py",
+  "codePathLocal":  "run_dpo.py",
+  "email":  "shaiksirajuddin9949@gmail.com",
+  "root":  "runs/dpo_run_24b_v1",
+  "host":  "a100-2gpu-shell-session-757d587799-mfdvv",
+  "executable":  "/workspace/llm_finetuning_env/bin/python",
+  "cpu_count":  12,
+  "cpu_count_logical":  24,
+  "gpu":  "NVIDIA A100-SXM4-80GB",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "791251738624",
+      "used":  "315579105280"
+    }
+  },
+  "memory":  {
+    "total":  "359047892992"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba"
+    },
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "05zaxpswkd8t9j7nmlszf02dmdnkyst5"
+}

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":2},"_runtime":2}

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-26T15:26:29.104421101Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpap_k1_i4/port-135896.txt","pid":135896,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2025-12-26T15:26:29.105138524Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":135896}
+{"time":"2025-12-26T15:26:29.105108641Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-135896-135976-1788857957/socket","Net":"unix"}}
+{"time":"2025-12-26T15:26:29.287192608Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2025-12-26T15:26:29.293378709Z","level":"INFO","msg":"handleInformInit: received","streamId":"ymohys1q","id":"1(@)"}
+{"time":"2025-12-26T15:26:29.445681764Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"ymohys1q","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.2673862Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.267537637Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-26T15:26:32.26754914Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.267646825Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.267756746Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-135896-135976-1788857957/socket","Net":"unix"}}
+{"time":"2025-12-26T15:26:32.914746116Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.914789004Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2025-12-26T15:26:32.914805725Z","level":"INFO","msg":"server is closed"}

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/logs/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:26:29.293501638Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:26:29.445455498Z","level":"INFO","msg":"stream: created new stream","id":"ymohys1q"}
+{"time":"2025-12-26T15:26:29.445544858Z","level":"INFO","msg":"handler: started","stream_id":"ymohys1q"}
+{"time":"2025-12-26T15:26:29.445672321Z","level":"INFO","msg":"stream: started","id":"ymohys1q"}
+{"time":"2025-12-26T15:26:29.445699121Z","level":"INFO","msg":"writer: started","stream_id":"ymohys1q"}
+{"time":"2025-12-26T15:26:29.445775229Z","level":"INFO","msg":"sender: started","stream_id":"ymohys1q"}
+{"time":"2025-12-26T15:26:32.267559653Z","level":"INFO","msg":"stream: closing","id":"ymohys1q"}
+{"time":"2025-12-26T15:26:32.786868259Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:26:32.913988671Z","level":"INFO","msg":"handler: closed","stream_id":"ymohys1q"}
+{"time":"2025-12-26T15:26:32.914088362Z","level":"INFO","msg":"sender: closed","stream_id":"ymohys1q"}
+{"time":"2025-12-26T15:26:32.914111382Z","level":"INFO","msg":"stream: closed","id":"ymohys1q"}

dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/logs/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_setup.py:_flush():80] Configure stats pid to 135896
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/logs/debug.log
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_152629-ymohys1q/logs/debug-internal.log
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:26:29,021 INFO    MainThread:135896 [wandb_init.py:init():889] starting backend
+2025-12-26 15:26:29,287 INFO    MainThread:135896 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:26:29,291 INFO    MainThread:135896 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:26:29,293 INFO    MainThread:135896 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:26:29,294 INFO    MainThread:135896 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:26:29,634 INFO    MainThread:135896 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:26:29,742 INFO    MainThread:135896 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:26:29,743 INFO    MainThread:135896 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:26:29,743 INFO    MainThread:135896 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:26:29,743 INFO    MainThread:135896 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:26:29,748 INFO    MainThread:135896 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:26:32,267 INFO    wandb-AsyncioManager-main:135896 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:26:32,267 INFO    wandb-AsyncioManager-main:135896 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            3md0aqv809sr0p87xotdtgert1x2be48:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "316388069376"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:29:02.292056Z"
+                writerId: 3md0aqv809sr0p87xotdtgert1x2be48
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/files/output.log

@@ -0,0 +1,44 @@
+Wandb initialized: project='dpo-training', name='auto-generated'
+2025-12-26 15:29:05,363 - INFO - Loading model config from ../../Models/Devstral-Small-2-24B-HS-CPT-SFT with trust_remote_code=True
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 560, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 960, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 751, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 574, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 560, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 960, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 751, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 574, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/files/requirements.txt

@@ -0,0 +1,104 @@
+exceptiongroup==1.3.1
+wheel==0.45.1
+python-dateutil==2.9.0.post0
+nvidia-ml-py==13.580.82
+huggingface_hub==1.2.3
+idna==3.11
+click==8.3.1
+numpy==2.2.6
+httpx==0.28.1
+tokenizers==0.22.1
+sympy==1.13.1
+yarl==1.22.0
+async-timeout==5.0.1
+datasets==4.4.2
+platformdirs==4.5.1
+nvidia-cuda-cupti-cu12==12.1.105
+nvidia-nvtx-cu12==12.1.105
+smmap==5.0.2
+accelerate==1.12.0
+requests==2.32.5
+aiohttp==3.13.2
+bitsandbytes==0.49.0
+nvidia-cublas-cu12==12.1.3.1
+mpmath==1.3.0
+torchaudio==2.5.1+cu121
+nvidia-cuda-runtime-cu12==12.1.105
+typing-inspection==0.4.2
+GitPython==3.1.45
+xxhash==3.6.0
+nvidia-cusolver-cu12==11.4.5.107
+pydantic_core==2.41.5
+six==1.17.0
+torchvision==0.20.1+cu121
+typing_extensions==4.15.0
+triton==3.1.0
+charset-normalizer==3.4.4
+nvitop==1.6.1
+wandb==0.23.1
+regex==2025.11.3
+pip==25.3
+nvidia-cusparse-cu12==12.1.0.106
+pytz==2025.2
+Jinja2==3.1.6
+psutil==7.2.0
+pillow==12.0.0
+packaging==25.0
+safetensors==0.7.0
+sentry-sdk==2.48.0
+gitdb==4.0.12
+httpcore==1.0.9
+setuptools==80.9.0
+nvidia-cufft-cu12==11.0.2.54
+anyio==4.12.0
+transformers==5.0.0.dev0
+pydantic==2.12.5
+fsspec==2025.10.0
+filelock==3.20.0
+PyYAML==6.0.3
+hf-xet==1.2.0
+nvidia-cudnn-cu12==9.1.0.70
+tqdm==4.67.1
+MarkupSafe==2.1.5
+attrs==25.4.0
+nvidia-cuda-nvrtc-cu12==12.1.105
+peft==0.18.0
+aiohappyeyeballs==2.6.1
+networkx==3.4.2
+nvidia-nvjitlink-cu12==12.9.86
+certifi==2025.11.12
+pyarrow==22.0.0
+dill==0.4.0
+protobuf==6.33.2
+aiosignal==1.4.0
+frozenlist==1.8.0
+urllib3==2.6.2
+propcache==0.4.1
+tzdata==2025.3
+pandas==2.3.3
+annotated-types==0.7.0
+shellingham==1.5.4
+nvidia-nccl-cu12==2.21.5
+multidict==6.7.0
+nvidia-curand-cu12==10.3.2.106
+trl==0.26.2
+torch==2.5.1+cu121
+h11==0.16.0
+multiprocess==0.70.18
+typer-slim==0.21.0
+wheel==0.45.1
+tomli==2.0.1
+autocommand==2.2.2
+jaraco.context==5.3.0
+zipp==3.19.2
+packaging==24.2
+inflect==7.3.1
+typing_extensions==4.12.2
+platformdirs==4.2.2
+jaraco.functools==4.0.1
+jaraco.collections==5.1.0
+jaraco.text==3.12.1
+backports.tarfile==1.2.0
+more-itertools==10.3.0
+importlib_metadata==8.0.0
+typeguard==4.3.0

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/files/wandb-metadata.json

@@ -0,0 +1,47 @@
+{
+  "os":  "Linux-6.12.46+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.10.12",
+  "startedAt":  "2025-12-26T15:29:02.292056Z",
+  "args":  [
+    "--config",
+    "config_dpo.yaml"
+  ],
+  "program":  "/workspace/trainer-kit/DPO/run_dpo.py",
+  "codePath":  "run_dpo.py",
+  "codePathLocal":  "run_dpo.py",
+  "email":  "shaiksirajuddin9949@gmail.com",
+  "root":  "runs/dpo_run_24b_v1",
+  "host":  "a100-2gpu-shell-session-757d587799-mfdvv",
+  "executable":  "/workspace/llm_finetuning_env/bin/python",
+  "cpu_count":  12,
+  "cpu_count_logical":  24,
+  "gpu":  "NVIDIA A100-SXM4-80GB",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "791251738624",
+      "used":  "316388069376"
+    }
+  },
+  "memory":  {
+    "total":  "359047892992"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba"
+    },
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "3md0aqv809sr0p87xotdtgert1x2be48"
+}

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":2},"_runtime":2}

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-26T15:29:02.378098877Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmph5oh7hv8/port-136944.txt","pid":136944,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2025-12-26T15:29:02.378909563Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":136944}
+{"time":"2025-12-26T15:29:02.378811163Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-136944-137024-1203613879/socket","Net":"unix"}}
+{"time":"2025-12-26T15:29:02.55980819Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2025-12-26T15:29:02.56654094Z","level":"INFO","msg":"handleInformInit: received","streamId":"spwc3b4a","id":"1(@)"}
+{"time":"2025-12-26T15:29:02.726927644Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"spwc3b4a","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.56156698Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.56167497Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.561719256Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-26T15:29:05.561768332Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.561970735Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-136944-137024-1203613879/socket","Net":"unix"}}
+{"time":"2025-12-26T15:29:05.964901945Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.964947772Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2025-12-26T15:29:05.964970982Z","level":"INFO","msg":"server is closed"}

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/logs/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:29:02.566705931Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:29:02.726675423Z","level":"INFO","msg":"stream: created new stream","id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:02.726786973Z","level":"INFO","msg":"handler: started","stream_id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:02.726918023Z","level":"INFO","msg":"stream: started","id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:02.726947115Z","level":"INFO","msg":"writer: started","stream_id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:02.726982198Z","level":"INFO","msg":"sender: started","stream_id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:05.56168481Z","level":"INFO","msg":"stream: closing","id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:05.886288482Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:29:05.963872485Z","level":"INFO","msg":"handler: closed","stream_id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:05.96397042Z","level":"INFO","msg":"sender: closed","stream_id":"spwc3b4a"}
+{"time":"2025-12-26T15:29:05.963978351Z","level":"INFO","msg":"stream: closed","id":"spwc3b4a"}

dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/logs/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_setup.py:_flush():80] Configure stats pid to 136944
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:29:02,293 INFO    MainThread:136944 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/logs/debug.log
+2025-12-26 15:29:02,294 INFO    MainThread:136944 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_152902-spwc3b4a/logs/debug-internal.log
+2025-12-26 15:29:02,294 INFO    MainThread:136944 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:29:02,294 INFO    MainThread:136944 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:29:02,294 INFO    MainThread:136944 [wandb_init.py:init():889] starting backend
+2025-12-26 15:29:02,560 INFO    MainThread:136944 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:29:02,564 INFO    MainThread:136944 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:29:02,566 INFO    MainThread:136944 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:29:02,567 INFO    MainThread:136944 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:29:02,898 INFO    MainThread:136944 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:29:03,007 INFO    MainThread:136944 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:29:03,007 INFO    MainThread:136944 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:29:03,007 INFO    MainThread:136944 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:29:03,007 INFO    MainThread:136944 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:29:03,012 INFO    MainThread:136944 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:29:05,561 INFO    wandb-AsyncioManager-main:136944 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:29:05,561 INFO    wandb-AsyncioManager-main:136944 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            b8zcal9au539h6npajcoix9vzmhwciop:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "317141377024"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:30:52.545366Z"
+                writerId: b8zcal9au539h6npajcoix9vzmhwciop
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/files/output.log

@@ -0,0 +1,44 @@
+Wandb initialized: project='dpo-training', name='auto-generated'
+2025-12-26 15:30:55,534 - INFO - Loading model config from ../../Models/Devstral-Small-2-24B-HS-CPT-SFT with trust_remote_code=True
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 560, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 960, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 751, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 574, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 560, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 960, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 751, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 574, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/files/requirements.txt

@@ -0,0 +1,104 @@
+exceptiongroup==1.3.1
+wheel==0.45.1
+python-dateutil==2.9.0.post0
+nvidia-ml-py==13.580.82
+huggingface_hub==1.2.3
+idna==3.11
+click==8.3.1
+numpy==2.2.6
+httpx==0.28.1
+tokenizers==0.22.1
+sympy==1.13.1
+yarl==1.22.0
+async-timeout==5.0.1
+datasets==4.4.2
+platformdirs==4.5.1
+nvidia-cuda-cupti-cu12==12.1.105
+nvidia-nvtx-cu12==12.1.105
+smmap==5.0.2
+accelerate==1.12.0
+requests==2.32.5
+aiohttp==3.13.2
+bitsandbytes==0.49.0
+nvidia-cublas-cu12==12.1.3.1
+mpmath==1.3.0
+torchaudio==2.5.1+cu121
+nvidia-cuda-runtime-cu12==12.1.105
+typing-inspection==0.4.2
+GitPython==3.1.45
+xxhash==3.6.0
+nvidia-cusolver-cu12==11.4.5.107
+pydantic_core==2.41.5
+six==1.17.0
+torchvision==0.20.1+cu121
+typing_extensions==4.15.0
+triton==3.1.0
+charset-normalizer==3.4.4
+nvitop==1.6.1
+wandb==0.23.1
+regex==2025.11.3
+pip==25.3
+nvidia-cusparse-cu12==12.1.0.106
+pytz==2025.2
+Jinja2==3.1.6
+psutil==7.2.0
+pillow==12.0.0
+packaging==25.0
+safetensors==0.7.0
+sentry-sdk==2.48.0
+gitdb==4.0.12
+httpcore==1.0.9
+setuptools==80.9.0
+nvidia-cufft-cu12==11.0.2.54
+anyio==4.12.0
+transformers==5.0.0.dev0
+pydantic==2.12.5
+fsspec==2025.10.0
+filelock==3.20.0
+PyYAML==6.0.3
+hf-xet==1.2.0
+nvidia-cudnn-cu12==9.1.0.70
+tqdm==4.67.1
+MarkupSafe==2.1.5
+attrs==25.4.0
+nvidia-cuda-nvrtc-cu12==12.1.105
+peft==0.18.0
+aiohappyeyeballs==2.6.1
+networkx==3.4.2
+nvidia-nvjitlink-cu12==12.9.86
+certifi==2025.11.12
+pyarrow==22.0.0
+dill==0.4.0
+protobuf==6.33.2
+aiosignal==1.4.0
+frozenlist==1.8.0
+urllib3==2.6.2
+propcache==0.4.1
+tzdata==2025.3
+pandas==2.3.3
+annotated-types==0.7.0
+shellingham==1.5.4
+nvidia-nccl-cu12==2.21.5
+multidict==6.7.0
+nvidia-curand-cu12==10.3.2.106
+trl==0.26.2
+torch==2.5.1+cu121
+h11==0.16.0
+multiprocess==0.70.18
+typer-slim==0.21.0
+wheel==0.45.1
+tomli==2.0.1
+autocommand==2.2.2
+jaraco.context==5.3.0
+zipp==3.19.2
+packaging==24.2
+inflect==7.3.1
+typing_extensions==4.12.2
+platformdirs==4.2.2
+jaraco.functools==4.0.1
+jaraco.collections==5.1.0
+jaraco.text==3.12.1
+backports.tarfile==1.2.0
+more-itertools==10.3.0
+importlib_metadata==8.0.0
+typeguard==4.3.0

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/files/wandb-metadata.json

@@ -0,0 +1,47 @@
+{
+  "os":  "Linux-6.12.46+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.10.12",
+  "startedAt":  "2025-12-26T15:30:52.545366Z",
+  "args":  [
+    "--config",
+    "config_dpo.yaml"
+  ],
+  "program":  "/workspace/trainer-kit/DPO/run_dpo.py",
+  "codePath":  "run_dpo.py",
+  "codePathLocal":  "run_dpo.py",
+  "email":  "shaiksirajuddin9949@gmail.com",
+  "root":  "runs/dpo_run_24b_v1",
+  "host":  "a100-2gpu-shell-session-757d587799-mfdvv",
+  "executable":  "/workspace/llm_finetuning_env/bin/python",
+  "cpu_count":  12,
+  "cpu_count_logical":  24,
+  "gpu":  "NVIDIA A100-SXM4-80GB",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "791251738624",
+      "used":  "317141377024"
+    }
+  },
+  "memory":  {
+    "total":  "359047892992"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba"
+    },
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "b8zcal9au539h6npajcoix9vzmhwciop"
+}

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":2},"_runtime":2}

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-26T15:30:52.629789303Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpvr96ff7k/port-138039.txt","pid":138039,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2025-12-26T15:30:52.630401044Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":138039}
+{"time":"2025-12-26T15:30:52.630403718Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-138039-138113-2700074418/socket","Net":"unix"}}
+{"time":"2025-12-26T15:30:52.815186583Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2025-12-26T15:30:52.821468691Z","level":"INFO","msg":"handleInformInit: received","streamId":"uo02exvi","id":"1(@)"}
+{"time":"2025-12-26T15:30:52.972985125Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"uo02exvi","id":"1(@)"}
+{"time":"2025-12-26T15:30:55.736694152Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2025-12-26T15:30:55.736777163Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2025-12-26T15:30:55.736810022Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-26T15:30:55.736839825Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2025-12-26T15:30:55.736962241Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-138039-138113-2700074418/socket","Net":"unix"}}
+{"time":"2025-12-26T15:30:56.144618955Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2025-12-26T15:30:56.144657804Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2025-12-26T15:30:56.144687588Z","level":"INFO","msg":"server is closed"}

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/logs/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:30:52.821600318Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:30:52.972768524Z","level":"INFO","msg":"stream: created new stream","id":"uo02exvi"}
+{"time":"2025-12-26T15:30:52.972861145Z","level":"INFO","msg":"handler: started","stream_id":"uo02exvi"}
+{"time":"2025-12-26T15:30:52.972976097Z","level":"INFO","msg":"stream: started","id":"uo02exvi"}
+{"time":"2025-12-26T15:30:52.973005528Z","level":"INFO","msg":"writer: started","stream_id":"uo02exvi"}
+{"time":"2025-12-26T15:30:52.973011764Z","level":"INFO","msg":"sender: started","stream_id":"uo02exvi"}
+{"time":"2025-12-26T15:30:55.736754973Z","level":"INFO","msg":"stream: closing","id":"uo02exvi"}
+{"time":"2025-12-26T15:30:56.052989718Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:30:56.143693278Z","level":"INFO","msg":"handler: closed","stream_id":"uo02exvi"}
+{"time":"2025-12-26T15:30:56.143802165Z","level":"INFO","msg":"sender: closed","stream_id":"uo02exvi"}
+{"time":"2025-12-26T15:30:56.143823041Z","level":"INFO","msg":"stream: closed","id":"uo02exvi"}

dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/logs/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:30:52,546 INFO    MainThread:138039 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:30:52,546 INFO    MainThread:138039 [wandb_setup.py:_flush():80] Configure stats pid to 138039
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/logs/debug.log
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_153052-uo02exvi/logs/debug-internal.log
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:30:52,547 INFO    MainThread:138039 [wandb_init.py:init():889] starting backend
+2025-12-26 15:30:52,815 INFO    MainThread:138039 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:30:52,819 INFO    MainThread:138039 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:30:52,821 INFO    MainThread:138039 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:30:52,822 INFO    MainThread:138039 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:30:53,141 INFO    MainThread:138039 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:30:53,249 INFO    MainThread:138039 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:30:53,249 INFO    MainThread:138039 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:30:53,250 INFO    MainThread:138039 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:30:53,250 INFO    MainThread:138039 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:30:53,255 INFO    MainThread:138039 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:30:55,736 INFO    wandb-AsyncioManager-main:138039 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:30:55,736 INFO    wandb-AsyncioManager-main:138039 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            3kcue4vill6aqei124wdyde6gjjhpcn4:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "317655887872"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:31:52.735494Z"
+                writerId: 3kcue4vill6aqei124wdyde6gjjhpcn4
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/files/output.log

@@ -0,0 +1,45 @@
+Wandb initialized: project='dpo-training', name='auto-generated'
+2025-12-26 15:31:55,903 - INFO - Loading model from ../../Models/Devstral-Small-2-24B-HS-CPT-SFT with trust_remote_code=True
+`torch_dtype` is deprecated! Use `dtype` instead!
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 559, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 957, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 748, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 572, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 559, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.
+
+During handling of the above exception, another exception occurred:
+
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 957, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 748, in main
+    model, tokenizer = load_base_model_and_tokenizer(cfg, base_dir)
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 572, in load_base_model_and_tokenizer
+    model = AutoModelForCausalLM.from_pretrained(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 376, in from_pretrained
+    raise ValueError(
+ValueError: Unrecognized configuration class <class 'transformers.models.mistral3.configuration_mistral3.Mistral3Config'> for this kind of AutoModel: AutoModelForCausalLM.
+Model type should be one of AfmoeConfig, ApertusConfig, ArceeConfig, AriaTextConfig, BambaConfig, BartConfig, BertConfig, BertGenerationConfig, BigBirdConfig, BigBirdPegasusConfig, BioGptConfig, BitNetConfig, BlenderbotConfig, BlenderbotSmallConfig, BloomConfig, BltConfig, CamembertConfig, LlamaConfig, CodeGenConfig, CohereConfig, Cohere2Config, CpmAntConfig, CTRLConfig, CwmConfig, Data2VecTextConfig, DbrxConfig, DeepseekV2Config, DeepseekV3Config, DiffLlamaConfig, DogeConfig, Dots1Config, ElectraConfig, Emu3Config, ErnieConfig, Ernie4_5Config, Ernie4_5_MoeConfig, Exaone4Config, FalconConfig, FalconH1Config, FalconMambaConfig, FlexOlmoConfig, FuyuConfig, GemmaConfig, Gemma2Config, Gemma3Config, Gemma3TextConfig, Gemma3nConfig, Gemma3nTextConfig, GitConfig, GlmConfig, Glm4Config, Glm4MoeConfig, GotOcr2Config, GPT2Config, GPT2Config, GPTBigCodeConfig, GPTNeoConfig, GPTNeoXConfig, GPTNeoXJapaneseConfig, GptOssConfig, GPTJConfig, GraniteConfig, GraniteMoeConfig, GraniteMoeHybridConfig, GraniteMoeSharedConfig, HeliumConfig, HunYuanDenseV1Config, HunYuanMoEV1Config, Jais2Config, JambaConfig, JetMoeConfig, Lfm2Config, Lfm2MoeConfig, LlamaConfig, Llama4Config, Llama4TextConfig, LongcatFlashConfig, MambaConfig, Mamba2Config, MarianConfig, MBartConfig, MegatronBertConfig, MiniMaxConfig, MinistralConfig, Ministral3Config, MistralConfig, MixtralConfig, MllamaConfig, ModernBertDecoderConfig, MoshiConfig, MptConfig, MusicgenConfig, MusicgenMelodyConfig, MvpConfig, NanoChatConfig, NemotronConfig, OlmoConfig, Olmo2Config, Olmo3Config, OlmoeConfig, OpenAIGPTConfig, OPTConfig, PegasusConfig, PersimmonConfig, PhiConfig, Phi3Config, Phi4MultimodalConfig, PhimoeConfig, PLBartConfig, ProphetNetConfig, Qwen2Config, Qwen2MoeConfig, Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig, RecurrentGemmaConfig, ReformerConfig, RemBertConfig, RobertaConfig, RobertaPreLayerNormConfig, RoCBertConfig, RoFormerConfig, RwkvConfig, SeedOssConfig, SmolLM3Config, StableLmConfig, Starcoder2Config, TrOCRConfig, VaultGemmaConfig, WhisperConfig, XGLMConfig, XLMConfig, XLMRobertaConfig, XLMRobertaXLConfig, XLNetConfig, xLSTMConfig, XmodConfig, ZambaConfig, Zamba2Config.

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/files/requirements.txt

@@ -0,0 +1,104 @@
+exceptiongroup==1.3.1
+wheel==0.45.1
+python-dateutil==2.9.0.post0
+nvidia-ml-py==13.580.82
+huggingface_hub==1.2.3
+idna==3.11
+click==8.3.1
+numpy==2.2.6
+httpx==0.28.1
+tokenizers==0.22.1
+sympy==1.13.1
+yarl==1.22.0
+async-timeout==5.0.1
+datasets==4.4.2
+platformdirs==4.5.1
+nvidia-cuda-cupti-cu12==12.1.105
+nvidia-nvtx-cu12==12.1.105
+smmap==5.0.2
+accelerate==1.12.0
+requests==2.32.5
+aiohttp==3.13.2
+bitsandbytes==0.49.0
+nvidia-cublas-cu12==12.1.3.1
+mpmath==1.3.0
+torchaudio==2.5.1+cu121
+nvidia-cuda-runtime-cu12==12.1.105
+typing-inspection==0.4.2
+GitPython==3.1.45
+xxhash==3.6.0
+nvidia-cusolver-cu12==11.4.5.107
+pydantic_core==2.41.5
+six==1.17.0
+torchvision==0.20.1+cu121
+typing_extensions==4.15.0
+triton==3.1.0
+charset-normalizer==3.4.4
+nvitop==1.6.1
+wandb==0.23.1
+regex==2025.11.3
+pip==25.3
+nvidia-cusparse-cu12==12.1.0.106
+pytz==2025.2
+Jinja2==3.1.6
+psutil==7.2.0
+pillow==12.0.0
+packaging==25.0
+safetensors==0.7.0
+sentry-sdk==2.48.0
+gitdb==4.0.12
+httpcore==1.0.9
+setuptools==80.9.0
+nvidia-cufft-cu12==11.0.2.54
+anyio==4.12.0
+transformers==5.0.0.dev0
+pydantic==2.12.5
+fsspec==2025.10.0
+filelock==3.20.0
+PyYAML==6.0.3
+hf-xet==1.2.0
+nvidia-cudnn-cu12==9.1.0.70
+tqdm==4.67.1
+MarkupSafe==2.1.5
+attrs==25.4.0
+nvidia-cuda-nvrtc-cu12==12.1.105
+peft==0.18.0
+aiohappyeyeballs==2.6.1
+networkx==3.4.2
+nvidia-nvjitlink-cu12==12.9.86
+certifi==2025.11.12
+pyarrow==22.0.0
+dill==0.4.0
+protobuf==6.33.2
+aiosignal==1.4.0
+frozenlist==1.8.0
+urllib3==2.6.2
+propcache==0.4.1
+tzdata==2025.3
+pandas==2.3.3
+annotated-types==0.7.0
+shellingham==1.5.4
+nvidia-nccl-cu12==2.21.5
+multidict==6.7.0
+nvidia-curand-cu12==10.3.2.106
+trl==0.26.2
+torch==2.5.1+cu121
+h11==0.16.0
+multiprocess==0.70.18
+typer-slim==0.21.0
+wheel==0.45.1
+tomli==2.0.1
+autocommand==2.2.2
+jaraco.context==5.3.0
+zipp==3.19.2
+packaging==24.2
+inflect==7.3.1
+typing_extensions==4.12.2
+platformdirs==4.2.2
+jaraco.functools==4.0.1
+jaraco.collections==5.1.0
+jaraco.text==3.12.1
+backports.tarfile==1.2.0
+more-itertools==10.3.0
+importlib_metadata==8.0.0
+typeguard==4.3.0

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/files/wandb-metadata.json

@@ -0,0 +1,47 @@
+{
+  "os":  "Linux-6.12.46+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.10.12",
+  "startedAt":  "2025-12-26T15:31:52.735494Z",
+  "args":  [
+    "--config",
+    "config_dpo.yaml"
+  ],
+  "program":  "/workspace/trainer-kit/DPO/run_dpo.py",
+  "codePath":  "run_dpo.py",
+  "codePathLocal":  "run_dpo.py",
+  "email":  "shaiksirajuddin9949@gmail.com",
+  "root":  "runs/dpo_run_24b_v1",
+  "host":  "a100-2gpu-shell-session-757d587799-mfdvv",
+  "executable":  "/workspace/llm_finetuning_env/bin/python",
+  "cpu_count":  12,
+  "cpu_count_logical":  24,
+  "gpu":  "NVIDIA A100-SXM4-80GB",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "791251738624",
+      "used":  "317655887872"
+    }
+  },
+  "memory":  {
+    "total":  "359047892992"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba"
+    },
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "3kcue4vill6aqei124wdyde6gjjhpcn4"
+}

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb":{"runtime":2},"_runtime":2}

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-26T15:31:52.82802196Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmpqkio4efp/port-138465.txt","pid":138465,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2025-12-26T15:31:52.828849471Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":138465}
+{"time":"2025-12-26T15:31:52.828808724Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-138465-138539-4238173929/socket","Net":"unix"}}
+{"time":"2025-12-26T15:31:53.011057052Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2025-12-26T15:31:53.017383764Z","level":"INFO","msg":"handleInformInit: received","streamId":"wxs32uu8","id":"1(@)"}
+{"time":"2025-12-26T15:31:53.176090883Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"wxs32uu8","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.103528174Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.103608964Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.103654139Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-26T15:31:56.103666347Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.103767972Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-138465-138539-4238173929/socket","Net":"unix"}}
+{"time":"2025-12-26T15:31:56.539431271Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.539481282Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2025-12-26T15:31:56.539506784Z","level":"INFO","msg":"server is closed"}

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/logs/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:31:53.017531144Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:31:53.175841849Z","level":"INFO","msg":"stream: created new stream","id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:53.175922557Z","level":"INFO","msg":"handler: started","stream_id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:53.176076095Z","level":"INFO","msg":"stream: started","id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:53.176100332Z","level":"INFO","msg":"writer: started","stream_id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:53.176110819Z","level":"INFO","msg":"sender: started","stream_id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:56.103613554Z","level":"INFO","msg":"stream: closing","id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:56.401489323Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:31:56.538246836Z","level":"INFO","msg":"handler: closed","stream_id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:56.538337251Z","level":"INFO","msg":"sender: closed","stream_id":"wxs32uu8"}
+{"time":"2025-12-26T15:31:56.538352318Z","level":"INFO","msg":"stream: closed","id":"wxs32uu8"}

dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/logs/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_setup.py:_flush():80] Configure stats pid to 138465
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/logs/debug.log
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_153152-wxs32uu8/logs/debug-internal.log
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:31:52,737 INFO    MainThread:138465 [wandb_init.py:init():889] starting backend
+2025-12-26 15:31:53,010 INFO    MainThread:138465 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:31:53,015 INFO    MainThread:138465 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:31:53,017 INFO    MainThread:138465 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:31:53,018 INFO    MainThread:138465 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:31:53,479 INFO    MainThread:138465 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:31:53,592 INFO    MainThread:138465 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:31:53,592 INFO    MainThread:138465 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:31:53,592 INFO    MainThread:138465 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:31:53,592 INFO    MainThread:138465 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:31:53,597 INFO    MainThread:138465 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:31:56,103 INFO    wandb-AsyncioManager-main:138465 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:31:56,103 INFO    wandb-AsyncioManager-main:138465 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            yg288fsgj0ia8gpi3lsxxd4iicxacpwh:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "318370951168"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:33:36.434359Z"
+                writerId: yg288fsgj0ia8gpi3lsxxd4iicxacpwh
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/files/output.log

@@ -0,0 +1,76 @@
+Wandb initialized: project='dpo-training', name='auto-generated'
+2025-12-26 15:33:39,399 - INFO - Detected Mistral3 model architecture, loading with specific class
+Loading weights: 100%|█| 585/585 [00:13<00:00, 41.90it/s, Materializing param=model.vision_tower.transfo
+2025-12-26 15:33:57,664 - INFO - Ensuring all parameters are materialized...
+Loading reference model (frozen copy)...
+2025-12-26 15:34:02,022 - INFO - Detected Mistral3 model architecture, loading with specific class
+Loading weights: 100%|█| 585/585 [00:13<00:00, 42.74it/s, Materializing param=model.vision_tower.transfo
+2025-12-26 15:34:17,738 - INFO - Ensuring all parameters are materialized...
+Reference model loaded and frozen
+2025-12-26 15:34:18,661 - INFO - HTTP Request: HEAD https://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/json/json.py "HTTP/1.1 200 OK"
+Generating train split: 7612 examples [00:00, 76349.43 examples/s]
+2025-12-26 15:34:18,785 - INFO - Formatting train DPO data...
+Formatting train DPO data (num_proc=4): 100%|██████████████| 6850/6850 [00:02<00:00, 2809.98 examples/s]
+Filter: 100%|█████████████████████████████████████████████| 6850/6850 [00:00<00:00, 58478.08 examples/s]
+2025-12-26 15:34:21,607 - INFO - Train dataset after filtering: 6850 examples
+2025-12-26 15:34:21,608 - INFO - train dataset validation passed: 6850 examples
+2025-12-26 15:34:21,608 - INFO - Formatting eval DPO data...
+Formatting eval DPO data (num_proc=4): 100%|██████████████████| 762/762 [00:02<00:00, 332.47 examples/s]
+Filter: 100%|███████████████████████████████████████████████| 762/762 [00:00<00:00, 36813.75 examples/s]
+2025-12-26 15:34:24,198 - INFO - Eval dataset after filtering: 762 examples
+2025-12-26 15:34:24,199 - INFO - eval dataset validation passed: 762 examples
+warmup_ratio is deprecated and will be removed in v5.2. Use `warmup_steps` instead.
+Early stopping enabled: patience=5, min_delta=0.001
+2025-12-26 15:34:24,239 - INFO - DPO Training with beta=0.1, loss_type=sigmoid
+warmup_ratio is deprecated and will be removed in v5.2. Use `warmup_steps` instead.
+Extracting prompt in train dataset: 100%|██████████████████| 6850/6850 [00:01<00:00, 5396.61 examples/s]
+Applying chat template to train dataset: 100%|█████████████| 6850/6850 [00:00<00:00, 8653.30 examples/s]
+Tokenizing train dataset:   0%|                                         | 0/6850 [00:00<?, ? examples/s]
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 1011, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 967, in main
+    trainer = DPOTrainer(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 480, in __init__
+    train_dataset = self._prepare_dataset(train_dataset, processing_class, args, "train")
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 654, in _prepare_dataset
+    dataset = dataset.map(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 562, in wrapper
+    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3341, in map
+    for rank, done, content in Dataset._map_single(**unprocessed_kwargs):
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3673, in _map_single
+    for i, example in iter_outputs(shard_iterable):
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3647, in iter_outputs
+    yield i, apply_function(example, i, offset=offset)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3570, in apply_function
+    processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 749, in process_row
+    processor, tokenizer = processing_class, processing_class.tokenizer  # the processing class is a processor
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 1331, in __getattr__
+    raise AttributeError(f"{self.__class__.__name__} has no attribute {key}")
+AttributeError: TokenizersBackend has no attribute tokenizer. Did you mean: '_tokenizer'?
+Traceback (most recent call last):
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 1011, in <module>
+    main()
+  File "/workspace/trainer-kit/DPO/run_dpo.py", line 967, in main
+    trainer = DPOTrainer(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 480, in __init__
+    train_dataset = self._prepare_dataset(train_dataset, processing_class, args, "train")
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 654, in _prepare_dataset
+    dataset = dataset.map(
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 562, in wrapper
+    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3341, in map
+    for rank, done, content in Dataset._map_single(**unprocessed_kwargs):
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3673, in _map_single
+    for i, example in iter_outputs(shard_iterable):
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3647, in iter_outputs
+    yield i, apply_function(example, i, offset=offset)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/datasets/arrow_dataset.py", line 3570, in apply_function
+    processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/trl/trainer/dpo_trainer.py", line 749, in process_row
+    processor, tokenizer = processing_class, processing_class.tokenizer  # the processing class is a processor
+  File "/workspace/llm_finetuning_env/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 1331, in __getattr__
+    raise AttributeError(f"{self.__class__.__name__} has no attribute {key}")
+AttributeError: TokenizersBackend has no attribute tokenizer. Did you mean: '_tokenizer'?

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/files/requirements.txt

@@ -0,0 +1,104 @@
+exceptiongroup==1.3.1
+wheel==0.45.1
+python-dateutil==2.9.0.post0
+nvidia-ml-py==13.580.82
+huggingface_hub==1.2.3
+idna==3.11
+click==8.3.1
+numpy==2.2.6
+httpx==0.28.1
+tokenizers==0.22.1
+sympy==1.13.1
+yarl==1.22.0
+async-timeout==5.0.1
+datasets==4.4.2
+platformdirs==4.5.1
+nvidia-cuda-cupti-cu12==12.1.105
+nvidia-nvtx-cu12==12.1.105
+smmap==5.0.2
+accelerate==1.12.0
+requests==2.32.5
+aiohttp==3.13.2
+bitsandbytes==0.49.0
+nvidia-cublas-cu12==12.1.3.1
+mpmath==1.3.0
+torchaudio==2.5.1+cu121
+nvidia-cuda-runtime-cu12==12.1.105
+typing-inspection==0.4.2
+GitPython==3.1.45
+xxhash==3.6.0
+nvidia-cusolver-cu12==11.4.5.107
+pydantic_core==2.41.5
+six==1.17.0
+torchvision==0.20.1+cu121
+typing_extensions==4.15.0
+triton==3.1.0
+charset-normalizer==3.4.4
+nvitop==1.6.1
+wandb==0.23.1
+regex==2025.11.3
+pip==25.3
+nvidia-cusparse-cu12==12.1.0.106
+pytz==2025.2
+Jinja2==3.1.6
+psutil==7.2.0
+pillow==12.0.0
+packaging==25.0
+safetensors==0.7.0
+sentry-sdk==2.48.0
+gitdb==4.0.12
+httpcore==1.0.9
+setuptools==80.9.0
+nvidia-cufft-cu12==11.0.2.54
+anyio==4.12.0
+transformers==5.0.0.dev0
+pydantic==2.12.5
+fsspec==2025.10.0
+filelock==3.20.0
+PyYAML==6.0.3
+hf-xet==1.2.0
+nvidia-cudnn-cu12==9.1.0.70
+tqdm==4.67.1
+MarkupSafe==2.1.5
+attrs==25.4.0
+nvidia-cuda-nvrtc-cu12==12.1.105
+peft==0.18.0
+aiohappyeyeballs==2.6.1
+networkx==3.4.2
+nvidia-nvjitlink-cu12==12.9.86
+certifi==2025.11.12
+pyarrow==22.0.0
+dill==0.4.0
+protobuf==6.33.2
+aiosignal==1.4.0
+frozenlist==1.8.0
+urllib3==2.6.2
+propcache==0.4.1
+tzdata==2025.3
+pandas==2.3.3
+annotated-types==0.7.0
+shellingham==1.5.4
+nvidia-nccl-cu12==2.21.5
+multidict==6.7.0
+nvidia-curand-cu12==10.3.2.106
+trl==0.26.2
+torch==2.5.1+cu121
+h11==0.16.0
+multiprocess==0.70.18
+typer-slim==0.21.0
+wheel==0.45.1
+tomli==2.0.1
+autocommand==2.2.2
+jaraco.context==5.3.0
+zipp==3.19.2
+packaging==24.2
+inflect==7.3.1
+typing_extensions==4.12.2
+platformdirs==4.2.2
+jaraco.functools==4.0.1
+jaraco.collections==5.1.0
+jaraco.text==3.12.1
+backports.tarfile==1.2.0
+more-itertools==10.3.0
+importlib_metadata==8.0.0
+typeguard==4.3.0

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/files/wandb-metadata.json

@@ -0,0 +1,47 @@
+{
+  "os":  "Linux-6.12.46+-x86_64-with-glibc2.35",
+  "python":  "CPython 3.10.12",
+  "startedAt":  "2025-12-26T15:33:36.434359Z",
+  "args":  [
+    "--config",
+    "config_dpo.yaml"
+  ],
+  "program":  "/workspace/trainer-kit/DPO/run_dpo.py",
+  "codePath":  "run_dpo.py",
+  "codePathLocal":  "run_dpo.py",
+  "email":  "shaiksirajuddin9949@gmail.com",
+  "root":  "runs/dpo_run_24b_v1",
+  "host":  "a100-2gpu-shell-session-757d587799-mfdvv",
+  "executable":  "/workspace/llm_finetuning_env/bin/python",
+  "cpu_count":  12,
+  "cpu_count_logical":  24,
+  "gpu":  "NVIDIA A100-SXM4-80GB",
+  "gpu_count":  2,
+  "disk":  {
+    "/":  {
+      "total":  "791251738624",
+      "used":  "318370951168"
+    }
+  },
+  "memory":  {
+    "total":  "359047892992"
+  },
+  "gpu_nvidia":  [
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba"
+    },
+    {
+      "name":  "NVIDIA A100-SXM4-80GB",
+      "memoryTotal":  "85899345920",
+      "cudaCores":  6912,
+      "architecture":  "Ampere",
+      "uuid":  "GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40"
+    }
+  ],
+  "cudaVersion":  "13.0",
+  "writerId":  "yg288fsgj0ia8gpi3lsxxd4iicxacpwh"
+}

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_runtime":49,"_wandb":{"runtime":49}}

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/logs/debug-core.log

@@ -0,0 +1,14 @@
+{"time":"2025-12-26T15:33:36.517152278Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmptn5phprh/port-138967.txt","pid":138967,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false}
+{"time":"2025-12-26T15:33:36.517842446Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":138967}
+{"time":"2025-12-26T15:33:36.517839815Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-138967-139051-1880805868/socket","Net":"unix"}}
+{"time":"2025-12-26T15:33:36.700377241Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"}
+{"time":"2025-12-26T15:33:36.707152235Z","level":"INFO","msg":"handleInformInit: received","streamId":"fb8js9es","id":"1(@)"}
+{"time":"2025-12-26T15:33:36.860726066Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"fb8js9es","id":"1(@)"}
+{"time":"2025-12-26T15:34:26.830462396Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"}
+{"time":"2025-12-26T15:34:26.830539704Z","level":"INFO","msg":"server is shutting down"}
+{"time":"2025-12-26T15:34:26.830534975Z","level":"INFO","msg":"connection: closing","id":"1(@)"}
+{"time":"2025-12-26T15:34:26.830623895Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"}
+{"time":"2025-12-26T15:34:26.830653617Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-138967-139051-1880805868/socket","Net":"unix"}}
+{"time":"2025-12-26T15:34:27.129538377Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"}
+{"time":"2025-12-26T15:34:27.129564503Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"}
+{"time":"2025-12-26T15:34:27.129572669Z","level":"INFO","msg":"server is closed"}

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/logs/debug-internal.log

@@ -0,0 +1,11 @@
+{"time":"2025-12-26T15:33:36.707301381Z","level":"INFO","msg":"stream: starting","core version":"0.23.1"}
+{"time":"2025-12-26T15:33:36.860475041Z","level":"INFO","msg":"stream: created new stream","id":"fb8js9es"}
+{"time":"2025-12-26T15:33:36.860568007Z","level":"INFO","msg":"handler: started","stream_id":"fb8js9es"}
+{"time":"2025-12-26T15:33:36.860715948Z","level":"INFO","msg":"stream: started","id":"fb8js9es"}
+{"time":"2025-12-26T15:33:36.860739836Z","level":"INFO","msg":"writer: started","stream_id":"fb8js9es"}
+{"time":"2025-12-26T15:33:36.860764109Z","level":"INFO","msg":"sender: started","stream_id":"fb8js9es"}
+{"time":"2025-12-26T15:34:26.830537011Z","level":"INFO","msg":"stream: closing","id":"fb8js9es"}
+{"time":"2025-12-26T15:34:27.027238553Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
+{"time":"2025-12-26T15:34:27.128722564Z","level":"INFO","msg":"handler: closed","stream_id":"fb8js9es"}
+{"time":"2025-12-26T15:34:27.12882072Z","level":"INFO","msg":"sender: closed","stream_id":"fb8js9es"}
+{"time":"2025-12-26T15:34:27.128830906Z","level":"INFO","msg":"stream: closed","id":"fb8js9es"}

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/logs/debug.log

@@ -0,0 +1,23 @@
+2025-12-26 15:33:36,435 INFO    MainThread:138967 [wandb_setup.py:_flush():80] Current SDK version is 0.23.1
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_setup.py:_flush():80] Configure stats pid to 138967
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_setup.py:_flush():80] Loading settings from /root/.config/wandb/settings
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_setup.py:_flush():80] Loading settings from /workspace/trainer-kit/DPO/wandb/settings
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_setup.py:_flush():80] Loading settings from environment variables
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_init.py:setup_run_log_directory():714] Logging user logs to runs/dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/logs/debug.log
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_init.py:setup_run_log_directory():715] Logging internal logs to runs/dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/logs/debug-internal.log
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_init.py:init():841] calling init triggers
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_init.py:init():846] wandb.init called with sweep_config: {}
+config: {'model': {'repo_id': '../../Models/Devstral-Small-2-24B-HS-CPT-SFT', 'revision': None, 'base_local_dir': 'base_model', 'trust_remote_code': True, 'tokenizer_use_fast': True, 'device_map': 'auto', 'torch_dtype': 'bfloat16', 'use_4bit': False, 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': False, 'bnb_4bit_compute_dtype': 'bfloat16', 'attn_implementation': None}, 'data': {'train_jsonl': 'dpo_pairs_generated.jsonl', 'eval_jsonl': None, 'eval_split_ratio': 0.1, 'prompt_field': 'prompt', 'chosen_field': 'chosen', 'rejected_field': 'rejected', 'score_field': 'f1_score', 'format_type': 'chatml', 'system_prompt': 'You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.\n\n## Output Format\n\n##OUTPUT\nExplain the data flow and why each component must change:\n- Flow: [Input → Processing → Output with arrows]\n- For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"\n- Explain coupling between components\n\n##SELECT\nmodify::crates/path/to/file.rs::impl::ComponentName\nadd::crates/another/file.rs::function::AnotherComponent\n<EOS>\n\n## Rules\n\n1. Use full paths: `remove::crates/folder/file.rs::Type::Name`\n2. Use `::` for nested items: `status::StructName::Type::Name`\n3. Always explain "must change because" and "without this"\n3. Types of components: function, struct, enum, impl, trait\n4. If there is extra information (e.g., enum variants), include that too.\n5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>\n', 'max_length': 2048, 'shuffle': True, 'num_proc': 4}, 'peft': {'enabled': True, 'r': 16, 'lora_alpha': 32, 'lora_dropout': 0.05, 'bias': 'none', 'target_modules': 'auto'}, 'dpo': {'beta': 0.1, 'label_smoothing': 0.0, 'loss_type': 'sigmoid', 'use_reference_model': True, 'reference_free': False}, 'train': {'num_train_epochs': 3, 'per_device_train_batch_size': 1, 'per_device_eval_batch_size': 1, 'gradient_accumulation_steps': 8, 'learning_rate': '5e-5', 'weight_decay': 0.0, 'warmup_ratio': 0.1, 'lr_scheduler_type': 'cosine', 'optim': 'adamw_torch', 'max_grad_norm': 1.0, 'gradient_checkpointing': True, 'logging_steps': 2, 'save_strategy': 'steps', 'save_steps': 100, 'save_total_limit': 10, 'evaluation_strategy': 'steps', 'eval_steps': 25, 'load_best_model_at_end': True, 'early_stopping': {'enabled': True, 'patience': 5, 'min_delta': 0.001, 'metric': 'eval_loss', 'mode': 'min'}, 'resume_from_checkpoint': 'auto'}, 'run_dir': 'runs/dpo_run_24b_v1', '_wandb': {}}
+2025-12-26 15:33:36,436 INFO    MainThread:138967 [wandb_init.py:init():889] starting backend
+2025-12-26 15:33:36,700 INFO    MainThread:138967 [wandb_init.py:init():892] sending inform_init request
+2025-12-26 15:33:36,705 INFO    MainThread:138967 [wandb_init.py:init():900] backend started and connected
+2025-12-26 15:33:36,708 INFO    MainThread:138967 [wandb_init.py:init():970] updated telemetry
+2025-12-26 15:33:36,709 INFO    MainThread:138967 [wandb_init.py:init():994] communicating run to backend with 90.0 second timeout
+2025-12-26 15:33:36,984 INFO    MainThread:138967 [wandb_init.py:init():1041] starting run threads in backend
+2025-12-26 15:33:37,095 INFO    MainThread:138967 [wandb_run.py:_console_start():2521] atexit reg
+2025-12-26 15:33:37,095 INFO    MainThread:138967 [wandb_run.py:_redirect():2369] redirect: wrap_raw
+2025-12-26 15:33:37,095 INFO    MainThread:138967 [wandb_run.py:_redirect():2438] Wrapping output streams.
+2025-12-26 15:33:37,095 INFO    MainThread:138967 [wandb_run.py:_redirect():2461] Redirects installed.
+2025-12-26 15:33:37,101 INFO    MainThread:138967 [wandb_init.py:init():1081] run started, returning control to user process
+2025-12-26 15:34:26,830 INFO    wandb-AsyncioManager-main:138967 [service_client.py:_forward_responses():80] Reached EOF.
+2025-12-26 15:34:26,830 INFO    wandb-AsyncioManager-main:138967 [mailbox.py:close():137] Closing mailbox, abandoning 1 handles.

dpo_run_24b_v1/wandb/run-20251226_153336-fb8js9es/run-fb8js9es.wandb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47b1b5cb59559e68f367989a74a0cbd677bc17c04c539a4bc2448b00f2fcb402
+size 410520

dpo_run_24b_v1/wandb/run-20251226_153517-g5bybskm/files/config.yaml

@@ -0,0 +1,165 @@
+_wandb:
+    value:
+        cli_version: 0.23.1
+        e:
+            47sn84x90gu1yvzd0dzhhujsd7q0za53:
+                args:
+                    - --config
+                    - config_dpo.yaml
+                codePath: run_dpo.py
+                codePathLocal: run_dpo.py
+                cpu_count: 12
+                cpu_count_logical: 24
+                cudaVersion: "13.0"
+                disk:
+                    /:
+                        total: "791251738624"
+                        used: "319211442176"
+                email: shaiksirajuddin9949@gmail.com
+                executable: /workspace/llm_finetuning_env/bin/python
+                gpu: NVIDIA A100-SXM4-80GB
+                gpu_count: 2
+                gpu_nvidia:
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-989794b0-ec3b-13bf-db9f-3fbe341497ba
+                    - architecture: Ampere
+                      cudaCores: 6912
+                      memoryTotal: "85899345920"
+                      name: NVIDIA A100-SXM4-80GB
+                      uuid: GPU-3790aa64-60ef-9eac-b0b1-b278ee8c0d40
+                host: a100-2gpu-shell-session-757d587799-mfdvv
+                memory:
+                    total: "359047892992"
+                os: Linux-6.12.46+-x86_64-with-glibc2.35
+                program: /workspace/trainer-kit/DPO/run_dpo.py
+                python: CPython 3.10.12
+                root: runs/dpo_run_24b_v1
+                startedAt: "2025-12-26T15:35:17.840098Z"
+                writerId: 47sn84x90gu1yvzd0dzhhujsd7q0za53
+        m: []
+        python_version: 3.10.12
+        t:
+            "1":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "2":
+                - 1
+                - 11
+                - 41
+                - 49
+                - 51
+                - 71
+                - 84
+                - 98
+            "3":
+                - 15
+                - 16
+            "4": 3.10.12
+            "5": 0.23.1
+            "6": 5.0.0.dev0
+            "12": 0.23.1
+            "13": linux-x86_64
+data:
+    value:
+        chosen_field: chosen
+        eval_jsonl: null
+        eval_split_ratio: 0.1
+        format_type: chatml
+        max_length: 2048
+        num_proc: 4
+        prompt_field: prompt
+        rejected_field: rejected
+        score_field: f1_score
+        shuffle: true
+        system_prompt: |
+            You are a Hyperswitch Rust code analyzer. Identify functions/structs that need modification for a given task.
+
+            ## Output Format
+
+            ##OUTPUT
+            Explain the data flow and why each component must change:
+            - Flow: [Input → Processing → Output with arrows]
+            - For each component: "The [ComponentName] ([path]) must [action] because [reason]—without this, [consequence]"
+            - Explain coupling between components
+
+            ##SELECT
+            modify::crates/path/to/file.rs::impl::ComponentName
+            add::crates/another/file.rs::function::AnotherComponent
+            <EOS>
+
+            ## Rules
+
+            1. Use full paths: `remove::crates/folder/file.rs::Type::Name`
+            2. Use `::` for nested items: `status::StructName::Type::Name`
+            3. Always explain "must change because" and "without this"
+            3. Types of components: function, struct, enum, impl, trait
+            4. If there is extra information (e.g., enum variants), include that too.
+            5. Start with ##OUTPUT, end with ##SELECT, terminate with <EOS>
+        train_jsonl: dpo_pairs_generated.jsonl
+dpo:
+    value:
+        beta: 0.1
+        label_smoothing: 0
+        loss_type: sigmoid
+        reference_free: false
+        use_reference_model: true
+model:
+    value:
+        attn_implementation: null
+        base_local_dir: base_model
+        bnb_4bit_compute_dtype: bfloat16
+        bnb_4bit_quant_type: nf4
+        bnb_4bit_use_double_quant: false
+        device_map: auto
+        repo_id: ../../Models/Devstral-Small-2-24B-HS-CPT-SFT
+        revision: null
+        tokenizer_use_fast: true
+        torch_dtype: bfloat16
+        trust_remote_code: true
+        use_4bit: false
+peft:
+    value:
+        bias: none
+        enabled: true
+        lora_alpha: 32
+        lora_dropout: 0.05
+        r: 16
+        target_modules: auto
+run_dir:
+    value: runs/dpo_run_24b_v1
+train:
+    value:
+        early_stopping:
+            enabled: true
+            metric: eval_loss
+            min_delta: 0.001
+            mode: min
+            patience: 5
+        eval_steps: 25
+        evaluation_strategy: steps
+        gradient_accumulation_steps: 8
+        gradient_checkpointing: true
+        learning_rate: "5e-5"
+        load_best_model_at_end: true
+        logging_steps: 2
+        lr_scheduler_type: cosine
+        max_grad_norm: 1
+        num_train_epochs: 3
+        optim: adamw_torch
+        per_device_eval_batch_size: 1
+        per_device_train_batch_size: 1
+        resume_from_checkpoint: auto
+        save_steps: 100
+        save_strategy: steps
+        save_total_limit: 10
+        warmup_ratio: 0.1
+        weight_decay: 0