News

vLLM
docs.vllm.ai > projects > ascend > en > v0.23.0 > developer_guide > performance_and_debug > optimization_and_tuning.html

Optimization and Tuning

7+ hour, 34+ min ago  (504+ words) This guide aims to help users improve vLLM Ascend performance at the system level. It includes OS configuration, library optimization, deployment guide, and so on. Any feedback is welcome. Install vLLM and vLLM Ascend: Please follow the Installation Guide to…...

vLLM
docs.vllm.ai > en > latest > api > vllm > models > inkling

inkling

1+ day, 12+ hour ago  (92+ words) vLLM Inkling model configs for the text backbone and audio/vision towers. NVFP4 (ModelOpt) support for the Inkling mixture-of-experts. Text-only entry point ( inkling_model checkpoints). Top-level (multimodal) entry point. Bases: _TmlForCausalLMBase, SupportsMultiModal Builds the vision + audio towers on top of the shared text backbone....

vLLM
docs.vllm.ai > en > latest > api > vllm > model_executor > layers > fused_moe > router > bf16x3_router_gemm_cutedsl

bf16x3_router_gemm_cutedsl

1+ day, 10+ hour ago  (64+ words) vLLM CuteDSL BF16x3 router GEMM. Computes X @ W.T for BF16 X with shape [N, K] and FP32 router weights W with shape [M, K] by decomposing each FP32 weight value into three BF16 residual terms inside the kernel, then accumulating the three BF16 MMA results into…...

vLLM
docs.vllm.ai > en > latest > api > vllm > models > inkling > nvidia > attention

attention

1+ day, 12+ hour ago  (13+ words) vLLM Project the per-head relative branch r to per-distance logits....

vLLM
docs.vllm.ai > en > latest > api > vllm > models > inkling > nvidia > ops > fa4_rel_attention

fa4_rel_attention

1+ day, 12+ hour ago  (82+ words) vLLM Round the FA4 scheduling bound up to a power of two. Return the split-KV cap for FA4 with sheared relative bias. Paged varlen FA4 over the bound K/V cache with the Inkling relative bias. q is (num_tokens, num_heads, head_dim); key_cache / value_cache are the paged caches (num_blocks, block_size, num_kv_heads, head_dim); block_table is…...

vLLM
docs.vllm.ai > en > latest > api > vllm > distributed > kv_transfer > kv_connector > v1 > offloading > config

config

1+ day, 8+ hour ago  (30+ words) vLLM Translate vLLM KV cache metadata for native offloading backends. Translate vLLM configuration into the native offloading boundary. Return whether a KV cache tensor uses a packed block stride....

vLLM
docs.vllm.ai > en > latest > api > vllm > models > inkling > nvidia > ops > mm_towers

mm_towers

1+ day, 12+ hour ago  (77+ words) vLLM Fused CUDA kernels for the Inkling vision/audio towers. Both kernels keep the reference paths' fp32 accumulation and per-op bf16 rounding points (native rms_norm / F.gelu ); outputs are frequently bit-identical and otherwise differ by 1-2 bf16 ulps from reduction-order (real-checkpoint-weight cosine vs reference > 0.9999998). Fused gelu…...

vLLM
docs.vllm.ai > en > latest > api > vllm > models > inkling > nvidia > ops > norm

norm

1+ day, 11+ hour ago  (42+ words) vLLM Fused res = residual + delta; y = rmsnorm(res). Fused rmsnorm(embed_table[input_ids], weight) row gather + norm. add_rmsnorm(residual, delta, weight, eps) ¶ Returns (y, res); both are fresh tensors (cudagraph-friendly, no in-place update of the inputs)....

vLLM
docs.vllm.ai > en > latest > api > vllm > v1 > kv_offload > tiering > p2p > session > session

session

1+ week, 1+ day ago  (232+ words) P2PSession — bidirectional session combining client + server roles. Wire protocol is unchanged. Both sides advertise their NIXL metadata via ConnectMsg when their session is connected; the peer's ConnectMsg triggers transport.add_remote_peer; ConnectAckMsg confirms the peer received our ConnectMsg, after which queued outgoing messages…...

vLLM
docs.vllm.ai > en > latest > api > vllm > parser > kimi_k2

kimi_k2

1+ week, 1+ day ago  (42+ words) vLLM Kimi K2 parser for reasoning and tool calls. Kimi K2 tool call format:: The header before is Kimi's native tool call id. The function name is the final component before:N. Kimi K2 parser backed by the declarative parser engine....