Support Hugging Face models (#215)

* support hf direct models
2025-09-12 15:25:00 +08:00 · 2024-01-03 15:13:26 -08:00
parent 1d09c4fecd
commit a5d6d0436c
16 changed files with 654 additions and 27 deletions
--- a/llms/speculative_decoding/decoder.py
+++ b/llms/speculative_decoding/decoder.py
@@ -27,7 +27,11 @@ class Tokenizer:

    def encode(self, s: str) -> mx.array:
        return mx.array(
-            self._tokenizer(s, return_tensors="np", return_attention_mask=False,)[
+            self._tokenizer(
+                s,
+                return_tensors="np",
+                return_attention_mask=False,
+            )[
                "input_ids"
            ].squeeze(0)
        )