# Error using deepspeed for sftconfig **URL:** \ **Category:** DeepSpeed\ **Created:** [April 20, 2025, 1:26pm UTC](https://discuss.huggingface.co/t/error-using-deepspeed-for-sftconfig/151433 "2025-04-20T13:26:32Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![sai-santhosh](https://avatars.discourse-cdn.com/v4/letter/s/ed655f/32.png) [@sai-santhosh](https://discuss.huggingface.co/u/sai-santhosh)\ **Post date:** [April 20, 2025, 1:26pm UTC](https://discuss.huggingface.co/t/error-using-deepspeed-for-sftconfig/151433/1 "2025-04-20T13:26:32Z")
```python from trl import SFTConfig import wandb wandb.init(mode="disabled") # Configure training arguments training_args = SFTConfig( output_dir="qwen_sign_language_interpretation", # Directory to save the model max_steps=250, per_device_train_batch_size=1, # Batch size for training per_device_eval_batch_size=1, # Batch size for evaluation gradient_accumulation_steps=16, # Steps to accumulate gradients gradient_checkpointing=True, # Enable gradient checkpointing for memory efficiency # Optimizer and scheduler settings optim="adamw_torch_fused", # Optimizer type learning_rate=2e-4, # Learning rate for training lr_scheduler_type="constant", # Type of learning rate scheduler # Logging and evaluation logging_steps=10, # Steps interval for logging # eval_steps=10, # Steps interval for evaluation # eval_strategy="steps", # Strategy for evaluation save_strategy="steps", # Strategy for saving the model save_steps=20, # Steps interval for saving metric_for_best_model="train_loss", # Metric to evaluate the best model greater_is_better=False, # Whether higher metric values are better # Mixed precision and gradient settings bf16=True, # Use bfloat16 precision deepspeed = "/content/zero_stage3_offload_config.json", max_grad_norm=0.3, # Maximum norm for gradient clipping warmup_ratio=0.03, # Ratio of total steps for warmup report_to="none", # Reporting tool for tracking metrics # Gradient checkpointing settings # gradient_checkpointing_kwargs={"use_reentrant": False}, # Options for gradient checkpointing # Dataset configuration dataset_text_field="", # Text field in dataset dataset_kwargs={"skip_prepare_dataset": True}, # Additional dataset options max_seq_length=1024 # Maximum sequence length for input ) training_args.remove_unused_columns = False # Keep unused columns in dataset ``` this is my code , im getting the below error.please help me resolve it ![image](https://us1.discourse-cdn.com/hellohellohello/original/3X/a/4/a41c96b3c0a90f420f872fa565e3503e89f71e56.png) ---
**Author:** ![John6666](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/john6666/32/27664_2.png) [@John6666](https://discuss.huggingface.co/u/John6666)\ **Post date:** [April 21, 2025, 5:50am UTC](https://discuss.huggingface.co/t/error-using-deepspeed-for-sftconfig/151433/2 "2025-04-21T05:50:15Z")
Perhaps unresolved issue… > > > \### System Info > \`\`\`python > 2023-08-24 17:23:17.908613: W tensorflow/compiler/tf…2tensorrt/utils/py\_utils.cc:38\] TF-TRT Warning: Could not find TensorRT > \[2023-08-24 17:23:20,478\] \[INFO\] \[real\_accelerator.py:158:get\_accelerator\] Setting ds\_accelerator to cuda (auto detect) > WARNING:tensorflow:From /usr/local/lib/python3.10/dist-packages/transformers/commands/env.py:100: is\_gpu\_available (from tensorflow.python.framework.test\_util) is deprecated and will be removed in a future version. > Instructions for updating: > Use \`tf.config.list\_physical\_devices('GPU')\` instead. > 2023-08-24 17:23:29.664543: W tensorflow/core/common\_runtime/gpu/gpu\_bfc\_allocator.cc:47\] Overriding orig\_value setting because the TF\_FORCE\_GPU\_ALLOW\_GROWTH environment variable is set. Original config value was 0. > \`\`\` > Copy-and-paste the text below in your GitHub issue and FILL OUT the two last points. > > \- \`transformers\` version: 4.32.0 > \- Platform: Linux-5.15.109+-x86\_64-with-glibc2.35 > \- Python version: 3.10.12 > \- Huggingface\_hub version: 0.16.4 > \- Safetensors version: 0.3.3 > \- Accelerate version: 0.22.0 > \- Accelerate config: not found > \- PyTorch version (GPU?): 2.0.1+cu118 (True) > \- Tensorflow version (GPU?): 2.12.0 (True) > \- Flax version (CPU?/GPU?/TPU?): 0.7.2 (gpu) > \- Jax version: 0.4.14 > \- JaxLib version: 0.4.14 > \- Using GPU in script?: \ > \- Using distributed or parallel set-up in script?: \ > > \### Who can help? > > @pacman > > \### Information > > \- \[\] The official example scripts > \- \[X\] My own modified scripts > > \### Tasks > > \- \[\] An officially supported task in the \`examples\` folder (such as GLUE/SQuAD, ...) > \- \[X\] My own task or dataset (give details below) > > \### Reproduction > > 1. Load deepspeed config into json file > 2. Pass into TrainingArguments > 3. Get error > > Here is my code: > \`\`\`python > import json > > deepspeed\_config = { > "fp16": { > "enabled": True, > "loss\_scale": 0, > "loss\_scale\_window": 1000, > "initial\_scale\_power": 16, > "hysteresis": 2, > "min\_loss\_scale": 1 > }, > "optimizer": { > "type": "AdamW", > "params": { > "lr": "auto", > "betas": "auto", > "eps": "auto", > "weight\_decay": "auto" > } > }, > "scheduler": { > "type": "WarmupLR", > "params": { > "warmup\_min\_lr": "auto", > "warmup\_max\_lr": "auto", > "warmup\_num\_steps": "auto" > } > }, > "zero\_optimization": { > "stage": 2, > "offload\_optimizer": { > "device": "cpu", > "pin\_memory": True > }, > "allgather\_partitions": True, > "allgather\_bucket\_size": 2e8, > "overlap\_comm": True, > "reduce\_scatter": True, > "reduce\_bucket\_size": 2e8, > "contiguous\_gradients": True > }, > "steps\_per\_print": 100, > "gradient\_accumulation\_steps": "auto", > "gradient\_clipping": "auto", > "train\_batch\_size": "auto", > "train\_micro\_batch\_size\_per\_gpu": "auto", > "activation\_checkpointing": { > "partition\_activations": True, > "contiguous\_memory\_optimization": True > }, > "wall\_clock\_breakdown": False > } > > > config\_filename = "deepspeed\_config.json" > > with open(config\_filename, 'w') as f: > json.dump(deepspeed\_config, f) > > trainer = Trainer( > model=model, > train\_dataset=train\_data, > eval\_dataset=val\_data, > args=TrainingArguments( > num\_train\_epochs=num\_epochs, > evaluation\_strategy='steps' if val\_set\_size \> 0 else 'no', > save\_strategy='steps', > eval\_steps=eval\_steps if val\_set\_size \> 0 else None, > save\_steps=save\_steps, > output\_dir=output\_dir, > save\_total\_limit=save\_total\_limit, > load\_best\_model\_at\_end=True if val\_set\_size \> 0 else False, > deepspeed='./deepspeed\_config.json', > ), > data\_collator=DataCollatorForSeq2Seq(tokenizer, pad\_to\_multiple\_of=8, return\_tensors='pt', padding=True), > callbacks=\[print\_callback\] > ) > model.config.use\_cache = False > trainer.train(resume\_from\_checkpoint=resume\_from\_checkpoint) > model.save\_pretrained(output\_dir) > \`\`\` > Here is my error: > \`\`\` > \--------------------------------------------------------------------------- > TypeError Traceback (most recent call last) > \[\\](https://localhost:8080/#) in \() > 3 train\_dataset=train\_data, > 4 eval\_dataset=val\_data, > \----\> 5 args=TrainingArguments( > 6 num\_train\_epochs=num\_epochs, > 7 evaluation\_strategy='steps' if val\_set\_size \> 0 else 'no', > > 3 frames > \[/usr/local/lib/python3.10/dist-packages/transformers/deepspeed.py\](https://localhost:8080/#) in \_\_init\_\_(self, config\_file\_or\_dict) > 64 dep\_version\_check("accelerate") > 65 dep\_version\_check("deepspeed") > \---\> 66 super().\_\_init\_\_(config\_file\_or\_dict) > 67 > 68 > > TypeError: object.\_\_init\_\_() takes exactly one argument (the instance to initialize) > \`\`\` > \### Expected behavior > > Trainer loads and runs. > > PS. This is my first ever issue reported, I'm a domain scientist sorry if this isn't the normal way to report things. > I’m don’t have a fix, but you could use 0.15.x version of deepspeed. I don’t get this error in that version