Install
Please confirm you are human
This browser or connection looks automated. Press and continuously hold the control for 3 seconds to enable Google-hosted web results and, when separately allowed, AI-assisted answers.
A successful check enables 100 search requests. Interactive access does not authorize scraping, systematic collection, or reuse of search output.
Web
NewsPlease enter a web search for web results.
News
WebGLM-5.2 744B-A40B with 256xH100??? Vime
2+ mon, 1+ day ago (412+ words) This is the recommended 32-node, 256-H100 training example for GLM-5.2. The recipe uses the GLM-5.2 BF16 checkpoint for Megatron training and the FP8 checkpoint for vLLM rollout. It assumes two Hugging Face repositories will be available: For environment setup and dataset download, see…...
minicpmv
9+ mon, 3+ week ago (96+ words) Different versions of MiniCPMV use different visual encoders and LLMs, which is not conducive to the current integration logic of LoRA and bitsandbytes in vLLM. Therefore, it is necessary to separate them. The abstract class of MiniCPMV can only be inherited,…...
humming
4+ mon, 6+ day ago (35+ words) vLLM Apply Humming-quantized MoE computation using the standard kernel flow. apply(layer, x, topk_weights, topk_ids, shared_experts, shared_experts_input) ¶ Humming handles quantization, weight application, and reduction internally in the experts implementation....
layernorm
9+ mon, 3+ week ago (108+ words) RMS normalization for Gemma. Root mean square normalization. RMS Normalization with optional gating. This is a native PyTorch implementation that supports: - Standard RMS normalization - Group RMS normalization - Optional gating with SiLU activation Size of the hidden dimension Epsilon for numerical…...
hy_v4
15+ hour, 50+ min ago (164+ words) HY V4 (hy_v4) model — hardware-isolated entry point. HY V4 combines three architectural pieces: The package is organized like vllm.models.deepseek_v32: this module is the only public entry point and dispatches on the current platform, so registry entries never reach into a platform subpackage....
model
15+ hour, 50+ min ago (68+ words) Inference-only HY V4 model compatible with HuggingFace weights (NVIDIA). One HY V4 decoder layer: MLA attention plus a dense or MoE MLP. Whenconfig.enable_ihc is set the layer runs on hc_mult residual channels and each sub-block is wrapped by an HYV4HCLayer boundary; otherwise it uses…...
attention
15+ hour, 50+ min ago (434+ words) MLA attention and lightning indexer for HY V4 (NVIDIA). The per-head learnable sink is supported through.flashmla_sparse, which subclasses the platform's sparse MLA backend to forward attn_sink. The sink is applied by binding the sink-capable backend from.flashmla_sparse; if no backend on this platform…...
hy_v4_reasoning_parser
15+ hour, 50+ min ago (286+ words) Reasoning extraction for HYV4, on plain data. Count reasoning tokens (0 in no_think). The start token is injected at the END of the prompt, so it is normally absent from token_ids. Only skip a leading start token when it actually appears first (legacy / no…...
flashmla_sparse
15+ hour, 50+ min ago (254+ words) Sink-capable FlashMLA sparse backend for HY V4 (NVIDIA). HY V4 adds a per-head learnable attention sink on top of sparse MLA. The vendored FlashMLA kernels already accept an attn_sink argument, but vLLM's shared FLASHMLA_SPARSE backend neither advertises sink support nor forwards the tensor, so…...
nvidia
15+ hour, 50+ min ago (51+ words) MLA attention and lightning indexer for HY V4 (NVIDIA). Sink-capable FlashMLA sparse backend for HY V4 (NVIDIA). iHC (independent Hyper-Connections) layers for HY V4 (NVIDIA). Inference-only HY V4 model compatible with HuggingFace weights (NVIDIA). Dense FFN and MoE blocks for HY V4 (NVIDIA). Multi-token prediction…...