dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[fix] reshape_paged_cache slot_mapping int64→int32 (C++ data_ptr<int>要求)

Browse Source
This commit is contained in:
root
2026-08-18 07:41:38 +00:00
parent ac6d51c936
commit ab078989da
2 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ex_engine/python/xllm_ops.py
View File

@@ -149,9 +149,11 @@ def gelu_and_mul(input, output=None):
# --- Cache (xllm_cache.so) --- # --- Cache (xllm_cache.so) ---
# C++ reshape_paged_cache(slot_ids, keys, values, key_cache, value_cache) # C++ reshape_paged_cache(slot_ids, keys, values, key_cache, value_cache)
# — slot_ids FIRST (not last!) # — slot_ids FIRST (not last!)
# — slot_ids must be int32 (C++ uses data_ptr<int>), vllm passes int64
def reshape_and_cache(key, value, key_cache, value_cache, slot_mapping): def reshape_and_cache(key, value, key_cache, value_cache, slot_mapping):
"""Write KV to paged cache. C++ takes slot_ids as FIRST arg.""" """Write KV to paged cache. C++ takes slot_ids as FIRST arg, dtype=int32."""
return _get("xllm_cache").reshape_paged_cache(slot_mapping, key, value, slot_mapping_i32 = slot_mapping.to(torch.int32)
return _get("xllm_cache").reshape_paged_cache(slot_mapping_i32, key, value,
key_cache, value_cache) key_cache, value_cache)
# --- Attention (ix_moe_bridge.so) --- # --- Attention (ix_moe_bridge.so) ---

6
qwen3_6_scripts/ex_engine/python/xllm_ops.py
View File

@@ -149,9 +149,11 @@ def gelu_and_mul(input, output=None):
# --- Cache (xllm_cache.so) --- # --- Cache (xllm_cache.so) ---
# C++ reshape_paged_cache(slot_ids, keys, values, key_cache, value_cache) # C++ reshape_paged_cache(slot_ids, keys, values, key_cache, value_cache)
# — slot_ids FIRST (not last!) # — slot_ids FIRST (not last!)
# — slot_ids must be int32 (C++ uses data_ptr<int>), vllm passes int64
def reshape_and_cache(key, value, key_cache, value_cache, slot_mapping): def reshape_and_cache(key, value, key_cache, value_cache, slot_mapping):
"""Write KV to paged cache. C++ takes slot_ids as FIRST arg.""" """Write KV to paged cache. C++ takes slot_ids as FIRST arg, dtype=int32."""
return _get("xllm_cache").reshape_paged_cache(slot_mapping, key, value, slot_mapping_i32 = slot_mapping.to(torch.int32)
return _get("xllm_cache").reshape_paged_cache(slot_mapping_i32, key, value,
key_cache, value_cache) key_cache, value_cache)
# --- Attention (ix_moe_bridge.so) --- # --- Attention (ix_moe_bridge.so) ---