[Distributed] Did not find tokenizer at {tokenizer_path}

[kwen2501]

🐛 Describe the bug

torchrun --nproc-per-node 8 dist_run.py

known configs: ['13B', '30B', '34B', '70B', '7B', 'CodeLlama-7b-Python-hf', 'Mistral-7B', 'stories110M', 'stories15M', 'stories42M', 'Meta-Llama-3-70B', 'Meta-Llama-3-8B', 'Meta-Llama-3.1-70B-Tune', 'Meta-Llama-3.1-70B', 'Meta-Llama-3.1-8B-Tune', 'Meta-Llama-3.1-8B']
09-14 15:41:32.092 - dist_run:137 - Chat Model Config: TransformerArgs(block_size=2048, vocab_size=32000, n_layers=32, n_heads=32, dim=4096, hidden_dim=11008, n_local_heads=32, head_dim=128, rope_base=10000, norm_eps=1e-05, multiple_of=256, ffn_dim_multiplier=None, use_tiktoken=False, max_seq_length=8192, rope_scaling=None, n_stages=1, stage_idx=0)
[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/kw2501/local/torchchat/dist_run.py", line 277, in <module>
[rank0]:     main()
[rank0]:   File "/home/kw2501/local/torchchat/dist_run.py", line 139, in main
[rank0]:     tokenizer = _build_chat_tokenizer()
[rank0]:   File "/home/kw2501/local/torchchat/dist_run.py", line 94, in _build_chat_tokenizer
[rank0]:     tokenizer_args = TokenizerArgs.from_args(args)
[rank0]:   File "/home/kw2501/local/torchchat/torchchat/cli/builder.py", line 269, in from_args
[rank0]:     raise RuntimeError(f"did not find tokenizer at {tokenizer_path}")
[rank0]: RuntimeError: did not find tokenizer at /home/kw2501/.torchchat/model-cache/meta-llama/Meta-Llama-3-8B-Instruct/tokenizer.model

Cc: @lessw2020

Versions

main branch

[lessw2020]

this highlights a current weakness - atm we are using the tokenizer downloaded by chat, but the safetensor weights by HF.
Thus, to get setup for a model we have to run:
1 - python3 torchchat.py download llama3 to get the tokenizer
2 - model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct") to get the safetensor weights.

[lessw2020]

we can hook into the torchchat download.py to download the tokenizer, and then pull the safetensor weights via an HF call as well.