diff --git a/docs/cookbook/autoregressive/Tencent/Hy4-Preview.mdx b/docs/cookbook/autoregressive/Tencent/Hy4-Preview.mdx index 4254cfb1a..2b283f23f 100644 --- a/docs/cookbook/autoregressive/Tencent/Hy4-Preview.mdx +++ b/docs/cookbook/autoregressive/Tencent/Hy4-Preview.mdx @@ -1,6 +1,6 @@ --- title: Hy4-Preview -description: "Deploy Tencent Hy4-Preview with SGLang — launch recipes for the 760B-parameter Mixture-of-Experts model with MLA, DeepSeek Sparse Attention (DSA), and MTP speculative decoding, in BF16 on H200/B200/B300/GB300 and MXFP8 on Blackwell GPUs." +description: "Deploy Tencent Hy4-Preview with SGLang — launch recipes for the 770B-parameter Mixture-of-Experts model with MLA, DeepSeek Sparse Attention (DSA), and MTP speculative decoding, in BF16 on H200/B200/B300/GB300 and MXFP8 on Blackwell GPUs." tag: NEW --- @@ -10,7 +10,7 @@ tag: NEW -For all methods and hardware platforms, see the [official SGLang installation guide](/docs/get-started/install). The two paths below match the **Python / Docker** toggle in the command panel. +For all methods and hardware platforms, see the [official SGLang installation guide](/docs/get-started/install). The Docker path below matches the **Docker** framing in the command panel. @@ -50,7 +50,7 @@ import { benchmarks } from "/src/snippets/configs/tencent/hy4-preview-benchmarks
  • ⧉ Copy — copies the current command (with whichever framing is active) to your clipboard.
  • $ cURL — a sample request against localhost:30000 to confirm the server is up.
  • ⚙ Env — edits the placeholders (HOST_IP, PORT, HF_TOKEN, NODE_RANK, NODE0_IP) the command and cURL share. Persists in localStorage across cookbooks.
    • -
  • Badge — every Hy4-Preview recipe currently shows In Progress: the B300 MXFP8 and H200 BF16 anchors have been served end-to-end, the remaining cells are sized from them by VRAM and architecture, and each cell flips to green Verified as its end-to-end verification lands.
    • +
  • Badge — single-node recipes show green Verified (run end-to-end on real hardware); the 2-node BF16 recipes show In Progress and flip to Verified as their multi-node verification lands.
    • @@ -60,10 +60,10 @@ The Playground lets you turn on additional knobs on top of whichever Deploy cell The knobs come in two flavors: -- **Built-in SGLang features** — parallelism overrides (TP / DP-Attention), MoE backend + EP, reasoning / tool-call parsers, speculative-decoding presets, prefill/decode disaggregation, and HiCache tiers. -- **Hy4 specific** — `--reasoning-parser auto` / `--tool-call-parser auto` resolve to the suffix-aware `hunyuan` parsers and read the real structural-token strings from the tokenizer vocab at runtime. +- **Built-in SGLang features** — parallelism overrides (TP / DP-Attention), the MoE a2a backend (DeepEP, which sets EP = TP), reasoning / tool-call parsers, speculative-decoding presets, prefill/decode disaggregation, and HiCache tiers. +- **Hy4 specific** — `--reasoning-parser auto` / `--tool-call-parser auto` resolve to the suffix-aware Hunyuan parsers and read the real structural-token strings from the tokenizer vocab at runtime. -Lines highlighted **green** are added by your overrides; lines with **red strikethrough** were in the base recipe but stripped by an override. Parallelism combinations beyond the listed recipes (DP-Attention, DeepEP/EP, other TP degrees) are experimentation territory — any override flips the badge to **Not Verified** until the configuration is run end-to-end and submitted back. +Lines highlighted **green** are added by your overrides; lines with **red strikethrough** were in the base recipe but stripped by an override. Parallelism combinations beyond the listed recipes (DP-Attention, DeepEP, other TP degrees) are experimentation territory — any override flips the badge to **Not Verified** until the configuration is run end-to-end and submitted back. import { Playground } from "/src/snippets/_playground.jsx"; @@ -78,7 +78,7 @@ import { Playground } from "/src/snippets/_playground.jsx"; ## 1. Model Introduction -**Hy4-Preview** is Tencent's next-generation flagship Mixture-of-Experts language model: ~760B total parameters with ~40B active per token, pairing a DeepSeek-style MLA + sparse-attention stack with Hunyuan's own residual control, MoE routing, and attention gating. It is a **text-only** model (vocab 120,832; the endpoint rejects image input by design). +**Hy4-Preview** is Tencent's next-generation flagship Mixture-of-Experts language model: 770B total parameters with 49B active per token, pairing a DeepSeek-style MLA + sparse-attention stack with Hunyuan's own residual control, MoE routing, and attention gating. It is a **text-only** model (vocab 120,832; the endpoint rejects image input by design), released under **Apache-2.0**. **Key architecture:** @@ -86,7 +86,7 @@ import { Playground } from "/src/snippets/_playground.jsx"; - **MLA + DSA on every layer**: Multi-head Latent Attention (`q_lora_rank` 2048, `kv_lora_rank` 512, 192 nope + 64 rope head dims, `v_head_dim` 256) under DeepSeek Sparse Attention — indexer top-k 2048 with 32 index heads, indexers alternating full/shared (one full indexer every 4 layers), and an FP8 index cache. - **iHC residual control**: Hunyuan's own inter-layer residual gating (`enable_ihc`, `hc_mult` 4) with pre- and post-residual gate groups. Semantically distinct from DeepSeek-V4's mHC (no combination step, no Sinkhorn) — the implementations are not interchangeable. - **Gated MLA + attention sink**: element-wise attention output gating evaluated in fp32, plus a learnable per-head attention sink propagated through all attention backends. -- **MTP (NextN)**: one built-in multi-token-prediction draft layer (`model.mtp_layers.0`, present in both checkpoints) enabling speculative decoding out of the box. +- **MTP (NextN)**: one built-in multi-token-prediction draft layer (`model.mtp_layers.0`, 10B parameters with ~0.7B active, present in both checkpoints) enabling speculative decoding out of the box. - **Long context**: 1M max positions (RoPE theta 1e7). Size `--context-length` to your KV budget — the sizing table in §2 suggests 262,144 (131,072 on H200). **Available models:** @@ -94,9 +94,11 @@ import { Playground } from "/src/snippets/_playground.jsx"; - [tencent/Hy4-preview](https://huggingface.co/tencent/Hy4-preview) — BF16 (~1.5TB weights) - [tencent/Hy4-preview-FP8](https://huggingface.co/tencent/Hy4-preview-FP8) — MXFP8 (ModelOpt recipe, UE8M0 group-32 weight scales, dynamic activations; ~760GB weights) -**Recommended generation:** SGLang applies the checkpoint's `generation_config.json` defaults — don't hardcode sampling parameters in client code. Thinking depth is controlled per request via OpenAI-standard `reasoning_effort` (`none` / `low` / `high`; see §3.1). +**Recommended generation:** `temperature=0.9`, `top_p=1.0` (informational — SGLang applies the checkpoint's `generation_config.json` defaults; don't hardcode sampling parameters in client code). Thinking depth is controlled per request via `reasoning_effort` (defaults to `high`; `no_think` for direct responses — see §3.1). -**Special tokens.** The Hy4-Preview tokenizer's structural tokens are suffix-bearing (``, ``, ``, ``, ``). SGLang's `hunyuan` reasoning/tool-call parsers resolve the real token strings from the tokenizer vocab at runtime, so `--reasoning-parser auto --tool-call-parser auto` work out of the box — auto-detection resolves both to the `hunyuan` parsers. +**Resources:** [HuggingFace](https://huggingface.co/tencent/Hy4-preview) · [GitHub](https://github.com/Tencent-Hunyuan/Hy4-preview). + +**Special tokens.** The Hy4-Preview tokenizer's structural tokens are suffix-bearing (``, ``, ``, ``, ``). SGLang's Hunyuan reasoning/tool-call parsers resolve the real token strings from the tokenizer vocab at runtime, so `--reasoning-parser auto --tool-call-parser auto` work out of the box. ## 2. Configuration Tips @@ -143,13 +145,13 @@ On H200 (BF16, TP16 → ~95GB of weights per rank) the per-rank pool holds rough **DSA attention backend.** Every layer runs DeepSeek Sparse Attention; SGLang auto-selects the DSA backend for HYV4 (`--attention-backend dsa` with `flashmla_sparse` prefill/decode and an FP8 indexer cache over a bf16 KV pool), so the recipes don't pass attention flags. Override only with a kernel-specific reason. -**MXFP8 kernel stack.** The MXFP8 checkpoint self-describes via its ModelOpt `hf_quant_config` (dynamic activations, UE8M0 group-32 weight scales) — no `--quantization` flag needed; the recipes pin `--fp8-gemm-backend deep_gemm` with the `deep_gemm` MoE runner on B300/GB300 and the `triton` MoE runner on B200. The MXFP8 kernel path requires SM100+ (Blackwell); H200 (SM90) cannot serve the MXFP8 checkpoint — use BF16 there. +**MXFP8 kernel stack.** The MXFP8 checkpoint self-describes via its ModelOpt `hf_quant_config` (dynamic activations, UE8M0 group-32 weight scales) — no `--quantization` flag needed; the recipes pin `--moe-runner-backend deep_gemm --fp8-gemm-backend deep_gemm`, the validated HYV4 MXFP8 path (the runtime also defaults both to `deep_gemm` for HYV4 when the flags are left unset). The MXFP8 kernel path requires SM100+ (Blackwell); H200 (SM90) cannot serve the MXFP8 checkpoint — use BF16 there. **CUDA graph decode vs eager.** Decode CUDA-graph capture is on by default; long-duration soak validation of the graph path on Hy4 is still in progress. If you hit instability under long mixed agentic workloads, pass `--disable-cuda-graph` to fall back to eager decode (a restart recovers cleanly either way). **MTP (NextN) speculative decoding.** Both checkpoints ship one draft layer; the preset is `--speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-num-draft-tokens 4` (top-k 1). Speculative decoding reserves 4 draft-token slots per request, so the effective request budget is `prompt_tokens + max_tokens + 4 ≤ context length` — requests at the exact context boundary are rejected with the reservation accounted for. -**Fail-fast guardrails.** The model rejects pipeline parallelism and `--enable-prefill-cp` before allocation. The recipes run pure TP; EP, DP-Attention, and other TP degrees are Playground experimentation territory. +**Fail-fast guardrails.** The model rejects pipeline parallelism and `--enable-prefill-cp` before allocation. The recipes run pure TP; DP-Attention, DeepEP (which sets EP = TP at runtime), and other TP degrees are Playground experimentation territory. **Multi-node BF16.** The BF16 weights don't fit a single H200/B200/GB300 host, so those cells are 2-node TP recipes — run the generated command on every node (the panel injects `--nnodes 2 --node-rank --dist-init-addr`) and keep the weights on storage shared across ranks. @@ -161,7 +163,7 @@ On H200 (BF16, TP16 → ~95GB of weights per rank) the per-rank pool holds rough ### 3.1 Reasoning (`reasoning_effort`) -Hy4-Preview is a hybrid-thinking model driven by the OpenAI-standard `reasoning_effort` field (`none` / `low` / `high` — note `none`, not `no_think`). Invalid values (including float efforts) are rejected with HTTP 400. The Deploy recipes enable the reasoning parser (`--reasoning-parser auto`) so thinking is separated into `reasoning_content` and the final answer into `content`: +Hy4-Preview is a hybrid-thinking model: reasoning defaults to `high` (deep chain-of-thought — suits math, coding, and agentic tasks). The top-level OpenAI `reasoning_effort` field accepts the standard tiers (e.g. `high`); for direct responses, pass the model-specific `no_think` through `extra_body={"chat_template_kwargs": {"reasoning_effort": "no_think"}}`. The Deploy recipes enable the reasoning parser (`--reasoning-parser auto`) so thinking is separated into `reasoning_content` and the final answer into `content`: @@ -194,7 +196,7 @@ Pending update — will be captured verbatim from a live Hy4-Preview server. - + ```python Example from openai import OpenAI @@ -204,7 +206,7 @@ client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY") response = client.chat.completions.create( model="tencent/Hy4-preview-FP8", messages=[{"role": "user", "content": "Give me a one-line summary of relativity."}], - reasoning_effort="none", + extra_body={"chat_template_kwargs": {"reasoning_effort": "no_think"}}, max_tokens=256, ) @@ -223,7 +225,7 @@ Pending update — will be captured verbatim from a live Hy4-Preview server. ### 3.2 Tool Calling -Hy4-Preview emits tool calls through suffixed structural tokens with an `arg_key` / `arg_value` argument format; SGLang's `hunyuan` tool-call parser reassembles them into OpenAI-compatible `message.tool_calls` with schema-aware type coercion, for both streaming and non-streaming requests. The Deploy recipes enable both parsers together (`--reasoning-parser auto --tool-call-parser auto`) — the reasoning parser strips thinking tokens before the tool-call parser runs. +Hy4-Preview emits tool calls through suffixed structural tokens with an `arg_key` / `arg_value` argument format; SGLang's Hunyuan tool-call parser reassembles them into OpenAI-compatible `message.tool_calls` with schema-aware type coercion, for both streaming and non-streaming requests. The Deploy recipes enable both parsers together (`--reasoning-parser auto --tool-call-parser auto`) — the reasoning parser strips thinking tokens before the tool-call parser runs. Tool-call output is parsed, not grammar-constrained: `tool_choice: "required"` / named-function forcing is not enforced with structural-tag guided decoding on the current implementation. diff --git a/docs/src/snippets/configs/tencent/hy4-preview.jsx b/docs/src/snippets/configs/tencent/hy4-preview.jsx index df877f7f9..cc2705732 100644 --- a/docs/src/snippets/configs/tencent/hy4-preview.jsx +++ b/docs/src/snippets/configs/tencent/hy4-preview.jsx @@ -2,7 +2,7 @@ // see _deployment.jsx header for the field contract. // // Sizing (drives the TP/nodes choices below): -// ~760B total / ~40B active MoE. BF16 weights ≈ 1.5TB → TP16 on H200/B200 +// 770B total / 49B active MoE. BF16 weights ≈ 1.5TB → TP16 on H200/B200 // (2x8 multi-node) or TP8 on B300 (single 8-GPU node) / GB300 (2x4 multi-node // — GB300 hosts carry 4 GPUs). MXFP8 ≈ 760GB → TP4 on B300/GB300 (288GB), // TP8 on B200; the MXFP8 kernel path requires SM100+, and H200 (SM90) was @@ -12,10 +12,11 @@ // left is ~25GB/rank ≈ 260K tokens — size `--context-length` to the pool // (the page's sizing table suggests 131072 there). // -// Every cell carries `verificationStatus: "in-progress"`. When a recipe's -// end-to-end verification lands, REPLACE that line with `verified: true` — -// `verificationStatus` takes precedence over `verified` in the engine, so -// merely adding `verified: true` would leave the badge amber. +// Single-node recipes are `verified: true` (run end-to-end); the 2-node +// BF16 recipes still carry `verificationStatus: "in-progress"` — when one +// lands, REPLACE that line with `verified: true` (`verificationStatus` +// takes precedence over `verified` in the engine, so merely adding +// `verified: true` would leave the badge amber). export const config = { modelName: "Hy4-Preview", @@ -173,37 +174,19 @@ sgl-eval run gsm8k \\ // ----- Card 2: "MoE Parallelism" ----- // 256 routed + 1 shared experts, top-8 sigmoid routing. The recipes run - // the MoE under pure TP (deep_gemm runner on B300/GB300 MXFP8, triton on - // B200); DeepEP/EP are experimentation overrides. No MegaMoE option — - // its fused path is not wired for Hy4's sigmoid-scored, bounded-SwiGLU - // experts. + // the MoE under pure TP (deep_gemm runner on MXFP8 — the validated + // HYV4 path); DeepEP is an experimentation override. No + // EP knob: the runtime rewrites EP to TP for a2a-spanning backends + // (DeepEP), so a free EP degree would advertise a topology that never + // runs. No MegaMoE option — its fused path is not wired for Hy4's + // sigmoid-scored, bounded-SwiGLU experts. moe: { backend: { options: [ { id: null, label: "Inherited" }, - { id: "deepep", label: "DeepEP", flags: ["--moe-a2a-backend deepep"] }, + { id: "deepep", label: "DeepEP (EP = TP)", flags: ["--moe-a2a-backend deepep"] }, ], }, - ep: { label: "EP", values: [ - null, - 4, - { value: 8, - disable: [ - { when: { effTp: [4] }, - reason: "EP=8 needs TP ≥ 8 (TP must be divisible by the EP degree) — raise TP in the Attention card first." }, - { when: { hw: ["gb300"], nodes: ["single"] }, - reason: "GB300 hosts carry 4 GPUs — EP=8 needs Multi-Nodes (2×4)." }, - ] }, - { value: 16, - disable: [ - { when: { effTp: [4, 8] }, - reason: "EP=16 needs TP=16 — raise TP in the Attention card first." }, - { when: { nodes: ["single"] }, - reason: "EP=16 requires 16 ranks — switch the Deploy panel's Nodes to Multi-Nodes first." }, - { when: { hw: ["gb300"] }, - reason: "GB300 hosts carry 4 GPUs — 2 nodes provide only 8 ranks." }, - ] }, - ]}, }, // ----- Card 3: "Parsers" ----- @@ -316,7 +299,7 @@ sgl-eval run gsm8k \\ // ==================================================================== { match: { hw: "b300", variant: "default", quant: "mxfp8", strategy: "low-latency", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", @@ -335,7 +318,7 @@ sgl-eval run gsm8k \\ }, { match: { hw: "b300", variant: "default", quant: "mxfp8", strategy: "high-throughput", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", @@ -355,7 +338,7 @@ sgl-eval run gsm8k \\ // ==================================================================== { match: { hw: "b300", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", @@ -372,7 +355,7 @@ sgl-eval run gsm8k \\ }, { match: { hw: "b300", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", @@ -422,18 +405,16 @@ sgl-eval run gsm8k \\ }, // ==================================================================== - // B200 (192GB) × MXFP8 — TP8 single node (~95GB weights/rank). B200 - // uses the triton MoE runner (the deep_gemm fused-expert path targets - // the 288GB SM103 parts). + // B200 (192GB) × MXFP8 — TP8 single node (~95GB weights/rank). // ==================================================================== { match: { hw: "b200", variant: "default", quant: "mxfp8", strategy: "low-latency", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", "--tp 8", - "--moe-runner-backend triton", + "--moe-runner-backend deep_gemm", "--fp8-gemm-backend deep_gemm", "--reasoning-parser auto", "--tool-call-parser auto", @@ -447,12 +428,12 @@ sgl-eval run gsm8k \\ }, { match: { hw: "b200", variant: "default", quant: "mxfp8", strategy: "high-throughput", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", "--tp 8", - "--moe-runner-backend triton", + "--moe-runner-backend deep_gemm", "--fp8-gemm-backend deep_gemm", "--reasoning-parser auto", "--tool-call-parser auto", @@ -502,7 +483,7 @@ sgl-eval run gsm8k \\ // ==================================================================== { match: { hw: "gb300", variant: "default", quant: "mxfp8", strategy: "low-latency", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}", @@ -521,7 +502,7 @@ sgl-eval run gsm8k \\ }, { match: { hw: "gb300", variant: "default", quant: "mxfp8", strategy: "high-throughput", nodes: "single" }, - verificationStatus: "in-progress", + verified: true, env: [], flags: [ "--model-path {{MODEL_NAME}}",