From 42346c6b39077aeadd29147f7b783418d409127e Mon Sep 17 00:00:00 2001
From: kingbri <8082010+kingbri1@users.noreply.github.com>
Date: Fri, 9 May 2025 22:11:05 -0400
Subject: [PATCH] Sampling: Remove skip_special_tokens

This parameter is way too confusing and does not make sense in
the modern LLM space.

Change approved by all maintainers.

Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
---
 backends/exllamav2/model.py         | 1 -
 common/sampling.py                  | 5 -----
 sampler_overrides/sample_preset.yml | 3 ---
 3 files changed, 9 deletions(-)

diff --git a/backends/exllamav2/model.py b/backends/exllamav2/model.py
index 22200ec..c2a084b 100644
--- a/backends/exllamav2/model.py
+++ b/backends/exllamav2/model.py
@@ -1373,7 +1373,6 @@ class ExllamaV2Container(BaseModelContainer):
             min_new_tokens=params.min_tokens,
             gen_settings=gen_settings,
             stop_conditions=stop_conditions,
-            decode_special_tokens=not params.skip_special_tokens,
             filters=grammar_handler.filters,
             filter_prefer_eos=bool(grammar_handler.filters),
             return_probs=params.logprobs > 0,
diff --git a/common/sampling.py b/common/sampling.py
index fc9f9bc..7e1d810 100644
--- a/common/sampling.py
+++ b/common/sampling.py
@@ -215,11 +215,6 @@ class BaseSamplerRequest(BaseModel):
         examples=[False],
     )
 
-    skip_special_tokens: Optional[bool] = Field(
-        default_factory=lambda: get_default_sampler_value("skip_special_tokens", True),
-        examples=[True],
-    )
-
     logit_bias: Optional[Dict[int, float]] = Field(
         default_factory=lambda: get_default_sampler_value("logit_bias"),
         examples=[{"1": 10, "2": 50}],
diff --git a/sampler_overrides/sample_preset.yml b/sampler_overrides/sample_preset.yml
index 9225976..907bbdd 100644
--- a/sampler_overrides/sample_preset.yml
+++ b/sampler_overrides/sample_preset.yml
@@ -136,9 +136,6 @@ add_bos_token:
 ban_eos_token:
   override: false
   force: false
-skip_special_tokens:
-  override: true
-  force: false
 logit_bias:
   override:
   force: false