[Inference/NFC] Clean outdated inference tests and deprecated kernels (…

…#5159) * [inference/nfc] remove outdated inference tests * remove outdated kernel tests * remove deprecated triton kernels * remove imports from deprecated kernels
hpcaitech · Dec 5, 2023 · 7fe0364 · 7fe0364
1 parent 778e679
commit 7fe0364
Show file tree

Hide file tree

Showing 18 changed files with 0 additions and 2,543 deletions.
diff --git a/colossalai/kernel/triton/__init__.py b/colossalai/kernel/triton/__init__.py
@@ -8,24 +8,12 @@
 
 # There may exist import error even if we have triton installed.
 if HAS_TRITON:
-    from .context_attention import bloom_context_attn_fwd, llama_context_attn_fwd
-    from .copy_kv_cache_dest import copy_kv_cache_to_dest
     from .fused_layernorm import layer_norm
     from .gptq_triton import gptq_fused_linear_triton
-    from .int8_rotary_embedding_kernel import int8_rotary_embedding_fwd
-    from .smooth_attention import smooth_llama_context_attn_fwd, smooth_token_attention_fwd
     from .softmax import softmax
-    from .token_attention_kernel import token_attention_fwd
 
     __all__ = [
-        "llama_context_attn_fwd",
-        "bloom_context_attn_fwd",
         "softmax",
         "layer_norm",
-        "copy_kv_cache_to_dest",
-        "token_attention_fwd",
         "gptq_fused_linear_triton",
-        "int8_rotary_embedding_fwd",
-        "smooth_llama_context_attn_fwd",
-        "smooth_token_attention_fwd",
     ]