3 people c98a8d6648 [Bugfix] Shared host arena: do not require seal constants in fcntl (#1404) ...

python-build-standalone CPython, which the Studio runtimes use, is built
against old kernel headers and has no fcntl.F_ADD_SEALS or F_SEAL_*, so
DeepSeek-V4.1's shared expert arena failed at startup with AttributeError
even though the kernel supports sealing. Fall back to the Linux ABI values.

Co-authored-by: mac <mac@MacBook-xinyi.local>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

2026-11-06 23:36:50 +08:00

..

_sm70

…

assets

…

benchmarks

…

compilation

[Bugfix][Compile] Keep clones that change layout in the clone elimination pass (#1359)

2026-11-06 06:58:19 +08:00

config

[Bugfix][SM70] GDN prefill: use the prebuilt kernel when TileLang has no nvcc (#1400)

2026-11-06 21:57:50 +08:00

cute_utils

…

device_allocator

…

distributed

[Bugfix][KV Offload] Keep external prefix hits on the Mamba page grid (#1360)

2026-11-05 21:30:15 +08:00

engine

[Perf][Core][SM70] Partial prefix-cache hits for hybrid Mamba models; agent turns resume at the last message (#1297)

2026-10-29 01:33:08 +08:00

entrypoints

[Bugfix][Frontend] Answer rejected requests without a server traceback (#1399)

2026-11-06 21:48:46 +08:00

image

…

inputs

…

ir

…

kernels

…

logging_utils

…

lora

…

media

…

model_executor

[Bugfix][SM70] Stop compiling prefill kernels for every new prompt length (#1394)

2026-11-06 09:48:19 +08:00

models

[Perf][SM70] Flash-Next: concurrent MTP4 verification batches (#1369)

2026-11-05 22:45:28 +08:00

multimodal

[Bugfix] Composite transparent image inputs before RGB conversion (#1366)

2026-11-05 21:32:21 +08:00

parser

[Bugfix] Preserve streaming tool calls, errors and logprob metadata (#1317)

2026-11-06 06:58:00 +08:00

platforms

[Bugfix][SM70] Release config: Flash-Next host KV default, 27B at c8, release image (#1392)

2026-11-06 07:27:29 +08:00

plugins

…

profiler

…

ray

…

reasoning

…

renderers

[Model] DeepSeek-V4.1-Flash on 4x V100: 512K packed KV, faster prefill, API launch (#1277)

2026-10-27 23:33:35 +08:00

sm70_profiles

[Bugfix][Core] Restore partial prefix hits on long hybrid prompts (27B and Flash-Next) (#1401)

2026-11-06 22:30:53 +08:00

third_party

[Model] DeepSeek-V4.1-Flash on 4x V100: 512K packed KV, faster prefill, API launch (#1277)

2026-10-27 23:33:35 +08:00

tokenizers

[Bugfix][Frontend] Bound the pending window of incremental detokenization (#1354)

2026-11-05 21:29:42 +08:00

tool_parsers

[Bugfix] Preserve streaming tool calls, errors and logprob metadata (#1317)

2026-11-06 06:58:00 +08:00

tracing

…

transformers_utils

[Model] DeepSeek-V4.1-Flash on 4x V100: 512K packed KV, faster prefill, API launch (#1277)

2026-10-27 23:33:35 +08:00

triton_utils

…

usage

…

utils

[Bugfix] Shared host arena: do not require seal constants in fcntl (#1404)

2026-11-06 23:36:50 +08:00

v1

[Perf][Core] MRV2 warmup: cover every request count, mixed steps and a full chunk (#1402)

2026-11-06 23:07:00 +08:00

video

…

vllm_flash_attn

…

__init__.py

…

_aiter_ops.py

…

_custom_ops.py

…

_sm70_ops.py

…

_xpu_ops.py

…

collect_env.py

…

connections.py

…

diagnostics.py

…

env_override.py

…

envs_metadata.py

…

envs.py

[Bugfix] Keep the pynccl all-reduce out-of-place by default (#1284)

2026-10-28 13:43:01 +08:00

exceptions.py

…

forward_context.py

…

logger.py

…

logits_process.py

…

logprobs.py

…

model_inspection.py

…

outputs.py

[Bugfix] Preserve prompt logprobs when merging parallel DELTA outputs (#1318)

2026-11-05 21:29:25 +08:00

pooling_params.py

…

py.typed

…

runtime_resources.py

…

sampling_params.py

…

scalar_type.py

…

scripts.py

…

sequence.py

…

sm70_decode_trace.py

…

sm70_gguf_quality.py

[Profile] Add an opt-in SM70 round-cost ledger (#1254)

2026-10-24 00:45:45 +08:00

sm70_graph_observer.py

…

sm70_round_cost.py

[Profile] Add an opt-in SM70 round-cost ledger (#1254)

2026-10-24 00:45:45 +08:00

tasks.py

…

version.py

…