Add HYV3 causal LM support

[ASheng1019]

Hi @EricLBuehler! This adds HYV3 support for the Tencent HyV3 / Hy-MT2 family.

The loader now supports the main causal LM path used by tencent/Hy-MT2-30B-A3B, including Q/K RMSNorm, sigmoid MoE routing with expert-bias top-k selection, the shared expert path, HYV3 RoPE config formats, fp32 lm_head projection when requested by the checkpoint, and HYV3 native tool calls.

I tested tencent/Hy-MT2-30B-A3B locally on a GB10 with CUDA. Text quality looked normal on translation / instruction prompts, and tool calling produced OpenAI-compatible tool_calls from the model's native HYV3 tag format.

With a CUDA build using cudnn + flash-attn, one smoke run produced:

• TTFT: 0.83s
• Prompt: 90 tokens, 109.36 tok/s
• Decode: 51 tokens, 27.17 tok/s

A forced tool-call smoke test on the same model completed successfully as well:

• Total time: 3.98s
• Prompt: 210 tokens, 249.70 tok/s
• Decode: 81 tokens, 25.80 tok/s

The same architecture should theoretically load the BF16 tencent/Hy3 main causal LM path as well: its config and weight index match the HYV3 main 80-layer backbone. I could not validate that end-to-end because Hy3 is too large for the GB10 setup I have.

Two known Hy3 gaps:

• Hy3’s extra MTP layer is not wired up for speculative decoding yet.
• tencent/Hy3-FP8 is not supported as-published yet because it needs FP8 checkpoint scale tensor support.

Would you be able to help validate tencent/Hy3 BF16 on hardware large enough for the main causal LM path?

[github-actions]

Code Metrics Report

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Language              Files        Lines         Code     Comments       Blanks
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 C Header                 23         4454         3116          790          548
 CSS                       3          282          252            6           24
 CUDA                    121        25439        20835         1707         2897
 Dockerfile                1           35           19            9            7
 HTML                      2           27           27            0            0
 JavaScript                3          578          563           12            3
 Jinja2                    7          694          656            5           33
 JSON                     27        17989        17986            0            3
 Makefile                  1           18           16            0            2
 MDX                      36         6285            0         4616         1669
 Metal Shading Lan|       37        14394        11386         1136         1872
 PowerShell                1          657          571           31           55
 Python                  151        12539        10418          480         1641
 Shell                     3         1071          852          115          104
 Plain Text               53        10687            0         9209         1478
 TOML                     28         1389         1209           39          141
 TypeScript               11         1649         1410           66          173
 YAML                      3           25           23            2            0
─────────────────────────────────────────────────────────────────────────────────
 Jupyter Notebooks         3          122           83           23           16
 |- Markdown               1           60           30           22            8
 |- Python                 1          122          113            1            8
 (Total)                              304          226           46           32
─────────────────────────────────────────────────────────────────────────────────
 Markdown                273        12055            0         8981         3074
 |- BASH                  24          300          221           47           32
 |- Dockerfile             2           14           12            0            2
 |- JSON                   6          289          289            0            0
 |- PowerShell             1            1            1            0            0
 |- Python               135         7349         6120          306          923
 |- Rust                  62         3840         2849          394          597
 |- TOML                   7           92           78            0           14
 (Total)                            23940         9570         9728         4642
─────────────────────────────────────────────────────────────────────────────────
 Rust                    701       331371       296672         5815        28884
 |- Markdown             418         9947          452         8304         1191
 (Total)                           341318       297124        14119        30075
─────────────────────────────────────────────────────────────────────────────────
 Svelte                   19         1969         1826           51           92
 |- CSS                    1            4            4            0            0
 |- JavaScript            19          921          767           25          129
 (Total)                             2894         2597           76          221
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Total                  1507       466668       378856        42192        45620
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━