← back to Exo
undo diff
3c81845ab74fb25052b32babaa249420c84d60e9 · 2024-11-27 16:42:19 +0000 · Rory Clear
Files touched
M exo/inference/tinygrad/inference.py
Diff
commit 3c81845ab74fb25052b32babaa249420c84d60e9
Author: Rory Clear <roryclear.rc@gmail.com>
Date: Wed Nov 27 16:42:19 2024 +0000
undo diff
---
exo/inference/tinygrad/inference.py | 1 +
1 file changed, 1 insertion(+)
diff --git a/exo/inference/tinygrad/inference.py b/exo/inference/tinygrad/inference.py
index 24cddc92..afa9c5df 100644
--- a/exo/inference/tinygrad/inference.py
+++ b/exo/inference/tinygrad/inference.py
@@ -51,6 +51,7 @@ def build_transformer(model_path: Path, shard: Shard, model_size="8B", device=No
weights = load(str(model_path), shard)
weights = convert_from_huggingface(weights, model, MODEL_PARAMS[model_size]["args"]["n_heads"], MODEL_PARAMS[model_size]["args"]["n_kv_heads"])
weights = fix_bf16(weights)
+
with Context(BEAM=0):
# replace weights in model
load_state_dict(model, weights, strict=False, consume=False) # consume=True
← 1d1fa8c6 move to tinygrad_helpers
·
back to Exo
·
add --default-temp option to change sample temperature c2647764 →