[KV Cache] GLM-5.3-Flash: use the generic packed KV layout; CircularBufferSpec tail with speculative ring slots (#57169)

Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Signed-off-by: Dakai An <dakaian108@gmail.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
This commit is contained in:
Dakai An
2026-10-09 15:48:37 -04:00
committed by GitHub
co-authored by Yifan Qiao OpenAI Codex OpenAI Codex Lucas Wilkinson Claude Opus 5.5 Lucas Wilkinson
parent 3766a67a32
commit 02e55d2af3
83 changed files with 1280 additions and 1986 deletions
+2 -6
View File
@@ -95,7 +95,7 @@ class Gemma4Proposer(SpecDecodeBaseProposer):
cm.block_table_tensor = self._per_group_block_tables[gid][:batch_size]
else:
cm = common_attn_metadata
attn_metadata = attn_group.get_metadata_builder().build_for_drafting(
attn_metadata = attn_group.build_metadata_for_drafting(
common_attn_metadata=cm, draft_index=draft_index
)
per_group_attn_metadata.append(attn_metadata)
@@ -291,11 +291,7 @@ class Gemma4Proposer(SpecDecodeBaseProposer):
self.draft_attn_groups = list(attention_groups.values())
if self.draft_attn_groups:
self.kv_cache_gid = self.draft_attn_groups[0].kv_cache_group_id
self.block_size = (
self.draft_attn_groups[0]
.get_metadata_builder()
.kv_cache_spec.block_size
)
self.block_size = self.draft_attn_groups[0].kv_cache_spec.block_size
else:
self.kv_cache_gid = 0
self.block_size = kv_cache_config.kv_cache_groups[
+2 -15
View File
@@ -991,7 +991,7 @@ class SpecDecodeBaseProposer:
per_group_attn_metadata: list[object] = []
per_layer_attn_metadata: dict[str, object] = {}
for attn_group in self.draft_attn_groups:
attn_metadata = attn_group.get_metadata_builder().build_for_drafting(
attn_metadata = attn_group.build_metadata_for_drafting(
common_attn_metadata=common_attn_metadata, draft_index=draft_index
)
per_group_attn_metadata.append(attn_metadata)
@@ -1765,20 +1765,7 @@ class SpecDecodeBaseProposer:
attention_groups[backend_key].layer_names.append(layer_name)
self.draft_attn_groups = list(attention_groups.values())
if kernel_block_sizes is not None and 0 <= self.kv_cache_gid < len(
kernel_block_sizes
):
# Slot mappings are computed against the block table, which is
# stored at kernel-block granularity. Use the kernel block size
# rather than the KV cache manager's block size; the two differ
# when manager blocks are split for the attention kernel.
self.block_size = kernel_block_sizes[self.kv_cache_gid]
else:
self.block_size = (
self.draft_attn_groups[0]
.get_metadata_builder()
.kv_cache_spec.block_size
)
self.block_size = self.draft_attn_groups[0].kv_cache_spec.block_size
logger.debug("Using block size %d for drafting layers", self.block_size)
def _determine_batch_execution_and_padding(
+2 -6
View File
@@ -144,7 +144,7 @@ class Step3p5MTPProposer(EagleProposer):
cm.slot_mapping = sm
else:
cm = common_attn_metadata
attn_metadata = attn_group.get_metadata_builder().build_for_drafting(
attn_metadata = attn_group.build_metadata_for_drafting(
common_attn_metadata=cm,
draft_index=draft_index,
)
@@ -244,11 +244,7 @@ class Step3p5MTPProposer(EagleProposer):
self.draft_attn_groups = list(attention_groups.values())
if self.draft_attn_groups:
self.kv_cache_gid = self.draft_attn_groups[0].kv_cache_group_id
self.block_size = (
self.draft_attn_groups[0]
.get_metadata_builder()
.kv_cache_spec.block_size
)
self.block_size = self.draft_attn_groups[0].kv_cache_spec.block_size
else:
self.kv_cache_gid = 0
self.block_size = kv_cache_config.kv_cache_groups[