ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

CI(Core): filter TRL trainer/config sweep to actual submodules only

Browse source 
The trainer-discovery sweep tripped on TRL 0.x (cell HF=4.57.6+TRL<1)
and TRL 1.x (cell HF=latest+TRL=latest) with:

  AST FAIL trl.trainer.get_peft_config: no spec
  AST FAIL trl.trainer.get_quantization_config: no spec

TRL re-exports those as utility FUNCTIONS in trl.trainer.__init__.
Their names end with `_config` so my `endswith("_config")` filter
swept them up alongside real `*_config.py` submodules; importlib.util.
find_spec then returns None because they are not files on disk and
the AST stage records `no spec` -> failure.

Add `_is_real_submodule(qual_name)` that tests `find_spec().origin`
non-None and apply it to both `_trainer_files()` and
`_config_files()`. Re-exported utility functions are silently
filtered out -- they are NOT modules and unsloth's auto-discovery in
rl.py:patch_trl_rl_trainers does not pretend they are.

Note: rl.py:1939-1943 has the same `endswith("_trainer")` filter
without a submodule check; it gets away with it today only because
TRL has no public `<x>_trainer`-suffixed function exports. If TRL
ever adds one, the same gap appears upstream.

Cell HF=default+TRL=default succeeded on the previous run because
its TRL pin (resolved via pyproject) happens to ship a different
public surface that does not include the `get_*_config` re-exports.

Verified locally on TRL 0.25.1: 16/16 raw `_config` names are real
submodules; 0 non-module exports filtered. Filter is a no-op on
versions without the trap and a corrective skip on versions with it.
This commit is contained in:
Daniel Han 2026-05-07 08:32:49 +00:00
commit f8860add83
1 changed files with 25 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

27
.github/workflows/consolidated-tests-ci.yml vendored
View file

@ -962,20 +962,43 @@ jobs:
import trl.trainer
# Replicate rl.py:1939-1943 verbatim.
def _is_real_submodule(qual_name: str) -> bool:
"""True iff `qual_name` resolves to an importable submodule
with a file on disk (i.e. has a non-None find_spec().origin).
TRL re-exports utility FUNCTIONS into `trl.trainer.__init__`
whose names happen to end with `_config` (e.g.
`get_peft_config`, `get_quantization_config`). Without this
filter the `endswith` check below picks them up as if they
were submodules and the AST stage fails on `no spec`. The
same trap exists for `_trainer` (none today, but defensive).
"""
try:
spec = importlib.util.find_spec(qual_name)
except (ImportError, ValueError):
return False
return spec is not None and bool(getattr(spec, "origin", None))
# Replicate rl.py:1939-1943 verbatim, then filter to actual
# submodules so re-exported utility functions (e.g.
# `get_peft_config`) do not pollute the AST sweep.
def _trainer_files():
return [
x for x in dir(trl.trainer)
if x.islower()
and x.endswith("_trainer")
and x != "base_trainer"
and _is_real_submodule(f"trl.trainer.{x}")
]
def _config_files():
return [
x for x in dir(trl.trainer)
if x.islower() and x.endswith("_config")
if x.islower()
and x.endswith("_config")
and _is_real_submodule(f"trl.trainer.{x}")
]