qwen4exp: support draft-head-only GGUFs (unsloth layout) + fix draft-load regression

[TheArchitectit]

qwen4exp: support draft-head-only GGUFs (unsloth layout) + fix draft-load regression

Companion patch to #27836 (qwen4exp architecture + NextN/MTP speculative decoding).

Two fixes that were necessary to run --spec-type draft-mtp with the unsloth-shipped Qwen3.8-Flash-Next GGUFs on CPU:

Fix 1: draft-head-only GGUFs

PR #27836's qwen4exp loader requires hc_head_norm/hc_head_down/hc_head_up and the PLE block unconditionally. The unsloth GGUFs ship draft-head-only files that lack these tensors — the target carries output_hc_* in its shards; the draft pack has only nextn.shared_head_norm + the block's own ffn projections.

This patch adds mtp_only detection (n_layer_nextn > 0 && blk.0.hc_attn_norm absent):

• trunk hc_head_* tensors marked TENSOR_NOT_REQUIRED when mtp_only
• PLE block guarded behind !mtp_only
• trunk tensor loop skipped when mtp_only (layers 0..n_layer-1 absent)
• MTP graph falls back to shared_head_norm + hc_ffn_down/hc_ffn_up when the head's own mixer is absent (mirrors the fallback path in the reference implementation)
• nextn.hc_head_* marked TENSOR_NOT_REQUIRED; new nextn.shared_head_norm tensor loaded when present

Fix 2: draft-load regression

common/speculative.cpp:2585 loaded params.model.path (the target model path) instead of model_path (the draft model path) — the variable holding the user-specified -md path. This caused the 147 GB target model to be loaded again as the draft, consuming all RAM and failing. One-line fix.

Testing

Tested on a 48-core Haswell Xeon (E5-2660 v3) with Qwen3.8-Flash-Next (UD-Q4_K_XL target, 4-shard GGUF, ~147 GB mlock) and the native MTP head quantized to Q8_0 as the draft model.

# upstream binary + PR #27836 + this patch
--spec-type draft-mtp --spec-draft-n-max 4 -md mtp-qwen4exp-Q8_0draft.gguf

# 8-prompt diverse suite (ab_8prompt.py)
ON  (draft-mtp Q8_0): mean 4.28 t/s, acceptance 0.51–0.75, mean_len 3.0–4.0
OFF (stock twin):     mean 3.33 t/s
Δ: +29% mean, 2.0× peak (best prompt 6.62 t/s vs 3.33)

Acceptance on the upstream per-stream combiner is dramatically higher than the mean-pooled alternative (0.51–0.75 vs 0.18–0.20, same head/flags), confirming PR #27836's note that the combiner must run per hc stream.

Drafter quant ladder (same upstream binary, dn=4)

All rungs requantized from the native BF16 MTP head pack (bit-identical lineage), arm-tested with journal-verified loads:

Quant	Size	Suite t/s	Mean acc_len	vs stock
Q8_0	3.9 GB	4.28	2.25	+29%
Q4_K_M	2.8 GB	3.97	2.34	+19%
Q5_0	2.8 GB	3.95	2.40	+19%
Q3_K_M	2.2 GB	3.44	2.15	+3%
Q2_K	1.8 GB	3.44	2.07	+3%

Acceptance binds only below ~5 bits; every rung beats stock (3.33 t/s). Q8_0 is the practical pick (identical speed to F16 at half the RAM; F16 ties at 4.29 t/s with highest acceptance 2.51 but costs 7.3 GB).

Draft-n-max sweep (Q8_0)

dn	Suite t/s	Mean acc_len	vs stock
2	3.72	1.88	+12%
4	4.28	2.25	+29%
6	3.52	2.47	+6%
8	3.38	2.60	+2%

dn=4 optimal; higher dn costs more verify than acceptance recovers.

Files changed

• src/models/qwen4exp.cpp — mtp_only detection, optional tensors, graph fallback (+25 lines)
• common/speculative.cpp — draft-load path fix (1 line)

[ggml-gh-bot]

Hi @TheArchitectit, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• PR Template not respected: Please respect the template when creating a new pull request. Make sure to fill out all required sections.

• AI-generated content: While code is allowed to be generated by AI, please write the PR description and commit messages on your own without the help of AI.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.