Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Fix three main CI failures from the Mac adapter export and seq2seq PRs (#12354)

Daniel Han37489ddcf78a
diff --git a/studio/backend/core/export/export.py b/studio/backend/core/export/export.py
index 3b94d2922f5..fb0f7ce0fd4 100644
--- a/studio/backend/core/export/export.py
+++ b/studio/backend/core/export/export.py
@@ -1808,6 +1808,8 @@ class ExportBackend:
check = True,
capture_output = True,
text = True,
+ encoding = "utf-8",
+ errors = "replace",
)
if not os.path.exists(source_dir):
os.replace(clone, source_dir)
@@ -1866,7 +1868,14 @@ class ExportBackend:
env = os.environ.copy()
apply_token_to_child_env(env, normalize_token(hf_token))
logger.info(f"Converting adapter at '{save_directory}' to GGUF -> '{out_gguf}'")
- result = subprocess.run(cmd, env = env, capture_output = True, text = True)
+ result = subprocess.run(
+ cmd,
+ env = env,
+ capture_output = True,
+ text = True,
+ encoding = "utf-8",
+ errors = "replace",
+ )
if result.returncode != 0:
raise RuntimeError(
f"LoRA -> GGUF conversion failed (exit {result.returncode}): "
diff --git a/studio/frontend/src/features/export/constants.ts b/studio/frontend/src/features/export/constants.ts
index a97b82d49ae..93ab3576da4 100644
--- a/studio/frontend/src/features/export/constants.ts
+++ b/studio/frontend/src/features/export/constants.ts
@@ -398,7 +398,7 @@ export function adapterCompatibilityTip(
}
return (
base +
- "loads in transformers, vLLM, and Unsloth Studio on GPU machines, and " +
+ "loads in transformers, vLLM, and Unsloth on GPU machines, and " +
"is required for GGUF adapter export."
);
}
diff --git a/tests/test_fa2_fast_generate_bypass.py b/tests/test_fa2_fast_generate_bypass.py
index 713c214c197..1b3d4d69e60 100644
--- a/tests/test_fa2_fast_generate_bypass.py
+++ b/tests/test_fa2_fast_generate_bypass.py
@@ -218,6 +218,8 @@ def test_wrapper_dispatch_preserves_normalization_and_selects_expected_path():
"transformers_version": TRANSFORMERS_VERSION,
"_uses_flash_attention_for_generation": uses_flash_attention,
"_clear_generation_caches": clear_generation_caches,
+ # Qwen3-VL is multimodal, so the real helper answers False for it too.
+ "_is_text_seq2seq_config": lambda config: False,
}
fast_generate = _load_function("unsloth_base_fast_generate", namespace)
@@ -303,6 +305,8 @@ def test_flash_attention_fallback_pins_a_dynamic_cache():
"transformers_version": TRANSFORMERS_VERSION,
"_uses_flash_attention_for_generation": uses_flash_attention,
"_clear_generation_caches": clear_generation_caches,
+ # Qwen3-VL is multimodal, so the real helper answers False for it too.
+ "_is_text_seq2seq_config": lambda config: False,
}
fast_generate = _load_function("unsloth_base_fast_generate", namespace)