From 7ef8cde3c22a3bc2240353353caa21051a42ea90 Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Tue, 21 Apr 2026 22:39:35 -0700
Subject: [PATCH] Coerce TRL's tuple-cached _*_available flags to bool (#5129)

transformers >= 4.48's `_is_package_available(name)` returns a tuple
`(bool, version_or_None)`. TRL's `trl.import_utils` caches that tuple
directly in `_vllm_ascend_available`, `_llm_blender_available`,
`_deepspeed_available`, `_joblib_available`, etc. and the matching
`is_*_available()` accessors return the tuple unchanged. A non-empty
tuple is always truthy, so `if is_vllm_ascend_available():` (in
`trl/extras/vllm_client.py`) fires unconditionally and triggers
`from vllm_ascend.distributed.device_communicators.pyhccl import ...`,
which fails outside Huawei Ascend hosts and blocks
`from trl import GRPOConfig, GRPOTrainer`. The same shape blocks
`is_llm_blender_available()` -> `import llm_blender` in
`trl/trainer/judges.py`.

Add `fix_trl_vllm_ascend()` to `import_fixes.py` and call it from
`unsloth/__init__.py` before any `from .trainer import *` that would
eagerly `import trl`. The fix walks `trl.import_utils` once and coerces
every `_*_available` tuple to a bool; the existing accessors that just
return the cached value then naturally yield a bool and the `if`
checks behave.
---
 unsloth/__init__.py     |  3 +++
 unsloth/import_fixes.py | 26 ++++++++++++++++++++++++++
 2 files changed, 29 insertions(+)

diff --git a/unsloth/__init__.py b/unsloth/__init__.py
index 73a9ce8fef..52114bb544 100644
--- a/unsloth/__init__.py
+++ b/unsloth/__init__.py
@@ -137,6 +137,7 @@ from .import_fixes import (
     fix_vllm_aimv2_issue,
     check_vllm_torch_sm100_compatibility,
     fix_vllm_guided_decoding_params,
+    fix_trl_vllm_ascend,
     fix_vllm_pdl_blackwell,
     fix_triton_compiled_kernel_missing_attrs,
     patch_trunc_normal_precision_issue,
@@ -159,6 +160,7 @@ fix_vllm_aimv2_issue()
 # Check vLLM + torch < 2.9.0 + SM100 compatibility BEFORE importing vLLM
 check_vllm_torch_sm100_compatibility()
 fix_vllm_guided_decoding_params()
+fix_trl_vllm_ascend()
 fix_vllm_pdl_blackwell()
 fix_triton_compiled_kernel_missing_attrs()
 patch_trunc_normal_precision_issue()
@@ -179,6 +181,7 @@ del fix_xformers_performance_issue
 del fix_vllm_aimv2_issue
 del check_vllm_torch_sm100_compatibility
 del fix_vllm_guided_decoding_params
+del fix_trl_vllm_ascend
 del fix_vllm_pdl_blackwell
 del fix_triton_compiled_kernel_missing_attrs
 del patch_trunc_normal_precision_issue
diff --git a/unsloth/import_fixes.py b/unsloth/import_fixes.py
index ca44a0ce7e..6cebad7939 100644
--- a/unsloth/import_fixes.py
+++ b/unsloth/import_fixes.py
@@ -489,6 +489,32 @@ def fix_vllm_guided_decoding_params():
         )
 
 
+def fix_trl_vllm_ascend():
+    # transformers >= 4.48's `_is_package_available(name)` returns a tuple
+    # (bool, version_or_None). TRL caches that tuple in module-level
+    # `_*_available` flags and the matching `is_*_available()` accessors
+    # return the tuple directly. A non-empty tuple is always truthy, so
+    # `if is_X_available():` fires even when X is absent, triggering an
+    # unconditional `import X` that fails. The surfaced case is
+    # `vllm_ascend` (blocks `from trl import GRPOConfig, GRPOTrainer`
+    # outside Huawei Ascend hosts); `llm_blender`, `deepspeed`, `joblib`
+    # share the same shape. Coerce every tuple-cached flag in
+    # trl.import_utils to bool; the existing accessors that just return
+    # the cached value then naturally yield a bool.
+    if importlib.util.find_spec("trl") is None:
+        return
+    try:
+        import trl.import_utils as tiu
+    except Exception:
+        return
+    for attr in list(vars(tiu)):
+        if not (attr.startswith("_") and attr.endswith("_available")):
+            continue
+        cached = getattr(tiu, attr)
+        if isinstance(cached, tuple):
+            setattr(tiu, attr, bool(cached and cached[0]))
+
+
 def ignore_logger_messages():
     # Ignore Environment variable `HF_TOKEN` is set
     try: