[Fix] Make the Solar model constructible and runnable

[ch-wan]

This PR is part of a stack (oldest at bottom):

• [Feature] Add --attn-dp-size and deprecate --enable-dp-attention #41818
• [Refactor] Enter draft TP scopes by attention ownership and drop ModelRunner.tp_group #41817
• [Refactor] Add make_pp_layers so models stop handing their PP position down #41816
• [Refactor] Stop passing models' layers the placement they already read #41815
• [Refactor] Let FusedMoE's weight-loading helpers read the layer's MoE-TP rank #41814
• [Refactor] Keep only the TP and PP groups on the model runner #41813
• [Refactor] Drop model placement attributes and parameters nothing reads #41812
• [Refactor] Drop placement values nothing reads #41811
• -> [Fix] Make the Solar model constructible and runnable #41810
• [Fix] Size the prefill delayer's gather buffer to its TP group #41809
• [Fix] Pass the draft's attention ownership to DFLASH's eager LiLiCorr scope #41808
• [Fix] Shard MoE WNA16 and Quark INT4-FP8 weights by the MoE placement #41807
• [Fix] State the configured MoE-DP width in the weight-cache fingerprint #41806
• [Fix] Read TensorCast's WORLD placement under its current names #41805

Motivation

SolarForCausalLM cannot be built or run:

• SolarModel calls make_layers without the pipeline position. In that case make_layers returns only the layer list, which is then unpacked as (start_layer, end_layer, layers), so construction fails with too many values to unpack.
• The forward passes call self.pp_group(), which is a group rather than a function. The causal-LM forward also drops pp_proxy_tensors, so a later pipeline stage would never receive the previous stage's hidden states.
• The logits processor is built and called with another framework's signatures: LogitsProcessor(vocab_size, org_vocab_size, logit_scale) and (lm_head, hidden_states, forward_batch).

Modifications

• Build the layers of this pipeline stage and read self.pp_group as an attribute.
• Pass pp_proxy_tensors through SolarForCausalLM.forward.
• Build and call the logits processor the way other models do. Under --enable-dp-lm-head, the LM head sits on the attention-TP group, as the processor expects.
• Solar's backbone skip connections mix hidden states saved at an earlier layer into a later one. A pipeline split that puts the two on different stages is now rejected at construction, naming the layer. test_solar_pipeline_split.py checks this against the checkpoint's skip layout: PP 1 and 2 are accepted, PP 4 is rejected.

Accuracy Tests

H200, upstage/solar-pro-preview-instruct (--trust-remote-code), GSM8K 200 questions:

	main	this PR
--tp-size 2	fails at construction	0.905
--tp-size 2 --pp-size 2		0.895; greedy output identical to TP2 on 4 prompts × 64 tokens
--pp-size 4		rejected: layers 16 and 48 read skip states saved on the previous stage

test/registered/unit at this PR's head, compared with main: no new failures.

Speed Tests and Profiling

Not applicable.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): 🚫 Run #36772646799
Latest PR Test (Extra): 🚫 Run #36772646360
Latest PR Test (AMD ROCm 10): ❌ Run #36772646960