Skip to content

Qwen3 Coder 480B inference sharding changes for MaxText. #4052

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
copybara-service wants to merge 1 commit into
base: main
Choose a base branch
from
+10 −2
Open

Qwen3 Coder 480B inference sharding changes for MaxText. #4052

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions src/maxtext/configs/pyconfig.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -503,6 +503,14 @@ def _initialize_pydantic(argv: list[str] | None = None, **kwargs) -> MaxTextConf

pydantic_kwargs = _prepare_for_pydantic(raw_keys_dict)

# Resolve relative tokenizer_path against the config directory (fileset root on Borg)
if pydantic_kwargs.get("tokenizer_path"):
fileset_root = os.path.dirname(config_path)
candidate_path = os.path.join(fileset_root, pydantic_kwargs["tokenizer_path"])
if os.path.exists(candidate_path):
logger.info("Resolved tokenizer_path %s to %s under fileset root", pydantic_kwargs["tokenizer_path"], candidate_path)
pydantic_kwargs["tokenizer_path"] = candidate_path

if pydantic_kwargs.get("use_tokamax_splash") and pydantic_kwargs.get("use_jax_splash"):
raise ValueError("At most one of `use_tokamax_splash` and `use_jax_splash` can be set to True.")

Expand Down
4 changes: 2 additions & 2 deletions src/maxtext/inference/decode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@ def main(argv: Sequence[str]) -> None:
# Prefill
rng, rng_prefill = jax.random.split(rng) # Split RNG before calling prefill
for i in range(_NUM_STREAMS):
with jax.profiler.StepTraceAnnotation("prefill", stream=i):
with jax.profiler.StepTraceAnnotation("prefill", step_num=i):
prefill_result, first_token = engine.prefill(
params=params,
padded_tokens=tokens,
Expand Down Expand Up @@ -206,7 +206,7 @@ def main(argv: Sequence[str]) -> None:
sampled_tokens_list.append(_batch_first_result_token(first_token_list, batch_size))
for i in steps:
rng, rng_generate = jax.random.split(rng)
with jax.profiler.StepTraceAnnotation("generate", step=i):
with jax.profiler.StepTraceAnnotation("generate", step_num=i):
decode_state, sampled_tokens = engine.generate(params, decode_state, rng=rng_generate)

# Automatically deactivate profiler after profiler_steps steps
Expand Down
Loading