CPU fp16 model hack

sxu · facebook-github-bot · commit 0ed61062711e · 2025-07-22T12:37:25.000-07:00
Differential Revision: D78753121
diff --git a/examples/models/llama/runner/static_attention_io_manager.h b/examples/models/llama/runner/static_attention_io_manager.h
@@ -903,7 +903,7 @@ class StaticAttentionIOManager {
   }
 
   StaticAttentionIOConfig config_;
-  size_t input_pos_;
+  size_t input_pos_ = 0;
   StaticKVCache<CacheT, CacheAllocatorT> kCaches_;
   StaticKVCache<CacheT, CacheAllocatorT> vCaches_;
   std::unordered_map<size_t, StaticAttentionMask<MaskT, MaskAllocatorT>>

Original file line number	Diff line number	Diff line change
`@@ -903,7 +903,7 @@ class StaticAttentionIOManager {`
`903`	`903`	`}`
`904`	`904`
`905`	`905`	`StaticAttentionIOConfig config_;`
`906`		`- size_t input_pos_;`
	`906`	`+ size_t input_pos_ = 0;`
`907`	`907`	`StaticKVCache<CacheT, CacheAllocatorT> kCaches_;`
`908`	`908`	`StaticKVCache<CacheT, CacheAllocatorT> vCaches_;`
`909`	`909`	`std::unordered_map<size_t, StaticAttentionMask<MaskT, MaskAllocatorT>>`