What is HUB_MODEL_ID environment variable? How is it diffrenent in this context from MODEL_ID. Why does the Axolotl examample, use HUB_MODEL_ID=meta-llama/Meta-Llama-3-70B?

@peterschmidt85 HUB_MODEL_ID : Repository ID where the model will be pushed on Hugging Face Hub (format: username/repo-name)

I will remove the assignment HUB_MODEL_ID=meta-llama/Meta-Llama-3-70B and only use HUB_MODEL_ID as in TRL example.

Do you know which specific drivers are missing in dstack's default Docker image?

cc @un-def

@peterschmidt85 As far as I remember dstack's default Docker image was issuing this error. libnccl-net.so was not found.

lambda-cluster-node-001:1652:1652 [6] NCCL INFO NET/Plugin: Plugin load returned 2 : libnccl-net.so: cannot open shared object file: No such file or directory : when loading libnccl-net.so
lambda-cluster-node-001:1652:1652 [6] NCCL INFO NET/Plugin: Using internal network plugin.
lambda-cluster-node-001:1651:1651 [5] NCCL INFO cudaDriverVersion 12080
lambda-cluster-node-001:1651:1651 [5] NCCL INFO Bootstrap : Using eno1:172.26.135.50<0>
lambda-cluster-node-001:1651:1651 [5] NCCL INFO NET/Plugin: No plugin found (libnccl-net.so)

Why use separate pip install commands instead of a single pip install command with multiple packages?

What about using uv pip install in examples since we now recommend uv?

I will update with uv pip install

For such multi-line commands as accelerate launch, should we use - | syntax?

@peterschmidt85 Yes, we can use - | like below for every multi-line commands.

 - |
   accelerate launch \
     --config_file=examples/accelerate_configs/fsdp1.yaml \
     --main_process_ip=$DSTACK_MASTER_NODE_IP \
     --main_process_port=8008 \
     --machine_rank=$DSTACK_NODE_RANK \
     --num_processes=$DSTACK_GPUS_NUM \
     --num_machines=$DSTACK_NODES_NUM \
     trl/scripts/sft.py \
     --model_name $MODEL_ID \
     --dataset_name OpenAssistant/oasst_top1_2023-08-25 \
     --dataset_text_field="text" \
     --per_device_train_batch_size 1 \
     --per_device_eval_batch_size 1 \
     --gradient_accumulation_steps 4 \
     --learning_rate 2e-4 \
     --report_to wandb 

This would make copy/paste multi-line command to shell very easy during debugging.

Need to set environment variables which values aren't configured in YAML (HF_TOKEN, WANDB_API_KEY, etc).

Add a link to the Clusters guide too

@peterschmidt85 I have added link to clusters guide in Create Fleet section as below
"For more detials on how to use clusters with dstack, check the Clusters guide."

therefore I did not add in What's next? section.