From bd32b2fd66a3b1d5d0169b47f8a8f8ba9ce1b79c Mon Sep 17 00:00:00 2001 From: DoubleMathew Date: Wed, 2 Jul 2025 18:59:56 -0500 Subject: [PATCH] Update CSM for faster inference (no compile) (#2865) --- unsloth/models/loader.py | 1 + 1 file changed, 1 insertion(+) diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py index a95a54b59d..683f2b1872 100644 --- a/unsloth/models/loader.py +++ b/unsloth/models/loader.py @@ -561,6 +561,7 @@ class FastModel(FastBaseModel): raise RuntimeError("Unsloth: Cohere's Command model only works on transformers >= 4.50.0." + NIGHTLY) # Sesame elif "csm-1b" in lowered_model_name: + os.environ["UNSLOTH_COMPILE_DISABLE"] = "1" # Inference is too slow os.environ["UNSLOTH_DISABLE_STATIC_GENERATION"] = "1" # Sesame fails os.environ["UNSLOTH_FORCE_CUSTOM_DTYPE"] = \ "all;torch.float32;torch.float16;"\