Compare commits

...
Sign in to create a new pull request.

2 commits

Author SHA1 Message Date
pre-commit-ci[bot]
d0c2beda75 [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
2026-03-18 10:34:15 +00:00
Daniel Han
798e73e770 fix(studio): add tiktoken to .venv_t5 for Qwen tokenizers
Qwen-family models (Qwen3.5, etc.) use a tiktoken-based tokenizer.
When Studio routes these models through the transformers 5.x overlay
(.venv_t5), the tokenizer instantiation fails with:

  ValueError: Couldn't instantiate the backend tokenizer...
  You need to have sentencepiece or tiktoken installed

This happens because .venv_t5 was set up with only transformers and
huggingface_hub (both --no-deps). The main venv usually has tiktoken,
but on Windows it sometimes does not.

Fix: install tiktoken into .venv_t5 in both setup.sh and the runtime
fallback (_ensure_venv_t5_exists). Uses --no-deps in the runtime path
since regex/requests are always present in the main venv.
2026-03-18 10:33:45 +00:00
2 changed files with 10 additions and 1 deletions

View file

@ -223,7 +223,11 @@ def _ensure_venv_t5_exists() -> bool:
logger.warning(".venv_t5 not found at %s — installing at runtime", _VENV_T5_DIR)
os.makedirs(_VENV_T5_DIR, exist_ok = True)
for pkg in (f"transformers=={TRANSFORMERS_5_VERSION}", "huggingface_hub==1.3.0"):
for pkg in (
f"transformers=={TRANSFORMERS_5_VERSION}",
"huggingface_hub==1.3.0",
"tiktoken",
):
cmd = [
sys.executable,
"-m",

View file

@ -254,6 +254,11 @@ else
mkdir -p "$VENV_T5_DIR"
run_quiet "pip install transformers 5.x" pip install --target "$VENV_T5_DIR" --no-deps "transformers==5.3.0"
run_quiet "pip install huggingface_hub for t5" pip install --target "$VENV_T5_DIR" --no-deps "huggingface_hub==1.3.0"
# tiktoken is needed by Qwen-family tokenizers (and others). The main venv
# may already have it, but on Windows the main venv sometimes doesn't, and
# the tokenizer loads while .venv_t5 is prepended to sys.path. Install it
# here so it is always available.
run_quiet "pip install tiktoken for t5" pip install --target "$VENV_T5_DIR" "tiktoken"
echo "✅ Transformers 5.x pre-installed to $VENV_T5_DIR/"
# ── 7. WSL: pre-install GGUF build dependencies ──