use GPT2TokenizerFast by default

2025-09-11 12:24:53 +00:00 · 2024-03-29 18:46:28 +01:00 · 2024-03-29 18:46:28 +01:00 · 275a61aae6
commit 275a61aae6
parent dcfefc425a
1 changed files with 22 additions and 11 deletions
--- a/server/text_generation_server/models/flash_dbrx.py
+++ b/server/text_generation_server/models/flash_dbrx.py
@ -37,6 +37,17 @@ class FlashDbrx(FlashCausalLM):
        else:
            raise NotImplementedError("FlashDBRX is only available on GPU")

+        try:
+            tokenizer = GPT2TokenizerFast.from_pretrained(
+                model_id,
+                revision=revision,
+                padding_side="left",
+                truncation_side="left",
+                trust_remote_code=trust_remote_code,
+                use_fast=True,
+                from_slow=False,
+            )
+        except:
            try:
                tokenizer = AutoTokenizer.from_pretrained(
                    model_id,