The supported range lives in requirements/common.txt. CI pins one exact
version in requirements/test/*.in and the compiled requirements/test/*.txt.
A version change is one of two jobs, and they usually come as a pair:
- Bump the pin: a new Transformers release becomes the CI version and the upper bound. vLLM grows compat code for it.
- Raise the floor: the minimum moves up. That compat code, and every vendored config or processor upstream now ships, becomes dead weight. Every floor raise ends with an unvendoring pass.
Bump the pin
- Edit the pin line in each
requirements/test/*.inand*.txtand the upper bound inrequirements/common.txt. Past bumps edit only the pin line in the compiled files; seegit log -- requirements/common.txt. - Read the release notes for config validation changes, renamed auto-mapping formats, and removed attributes; these are what break vLLM.
- Search
tests/formin_transformers_versionand skip markers naming the new version, and run those tests on the new pin. Keep the gates: older versions in the supported range still lack the feature, so a gate goes only when the floor reaches it. - Run the suites in Testing and fix what breaks. Mark every
shim with a
# Transformers >= Xcomment so the next floor raise finds it.
Done when the Testing suites and every test gated on this version pass on the new pin.
Raise the floor (unvendoring pass)
1. Inventory
Inventory candidates with the floor version installed (no vLLM install needed):
uv run --no-project --with transformers==<floor> \
python .agents/skills/transformers-upgrade/scripts/find_upstreamed.py
It lists:
- vendored configs whose
model_typeupstream now owns; _CONFIG_REGISTRYoverrides of upstream types;- registered processors that collide with upstream class names or models;
- version-gated code at or below the floor;
- code that reads the installed Transformers version, whatever its threshold;
- patches citing a Transformers issue or PR (with whether a referenced PR is in the floor release) or a future minimum version.
It finds candidates by name only. Every hit still needs a verdict.
2. Port each candidate to upstream
Every hit is deleted by default. A vendored copy is a liability: it drifts from upstream, hides upstream fixes, and shadows the native class. Port the vLLM model code to the upstream schema instead, renaming attribute reads and deriving values from upstream fields.
Fan candidates out to parallel subagents. Each one compares the vendored and upstream classes, then greps every attribute the vLLM model code reads. Upstreaming hazards lists the usual fixes.
Keep a vendored class only when upstream cannot load a checkpoint vLLM must keep serving, and record why in the PR description. A different upstream schema, a model-code rewrite, or a needed eval is work for the PR, not a reason to keep.
3. Prove it on real checkpoints
For every checkpoint in tests/models/registry.py that uses a ported
config, dump what vLLM builds before and after the change, then diff the two.
Run both dumps; each catches what the other misses:
dump_configs.py: the resolved config, attribute by attribute.dump_model.py: the model built on the meta device with no GPU or weights. It records ModelConfig sizes, every parameter shape, and every module's scalar attributes (head_dim,num_kv_heads,scaling, ...). Missing fields that are read with defaults only show up here.
git worktree add --detach <scratch>/vllm-base upstream/main
S=.agents/skills/transformers-upgrade/scripts
for dump in dump_configs dump_model; do
PYTHONPATH=<scratch>/vllm-base .venv/bin/python $S/$dump.py before.jsonl <repos>
PYTHONPATH=$PWD .venv/bin/python $S/$dump.py after.jsonl <repos>
.venv/bin/python $S/diff_dumps.py before.jsonl after.jsonl
done
Add --trust-remote-code for checkpoints whose config.json has an
auto_map, and dump both ways.
How to read the diffs:
- A
<MISSING>config field the model reads: this crashes or silently falls back to a default. Port the model code to the upstream field. - New upstream-only defaults (
attention_bias=False,pad_token_id=None): usually benign. A model diff of 0 confirms it. - Any model diff: this is a numerics change, and the PR needs a model
eval. For example, A.X-K1 gained the checkpoint's YaRN rope scaling, which
the vendored class had been dropping; it shows up as
layers.*.self_attn.scaling (x61).
4. Delete
For each deleted config:
- Delete the file.
- Remove its entries from
_CLASS_TO_MODULEand__all__invllm/transformers_utils/configs/__init__.py. - Remove its entries from
_CONFIG_REGISTRYinvllm/transformers_utils/config.py. - Switch every importer, including tests, to
from transformers import .... - Remove the
source_file_dependencieslines naming the file from.buildkite/.
Deleted processors follow the same steps in
vllm/transformers_utils/processors/__init__.py.
Before merging, rebase and grep again for the old field names: PRs that land in between still read them.
Done when every line of the inventory output is deleted, or kept with a reason that meets the bar in step 2. The PR also carries the config diff results and evals for any changed values.
Upstreaming hazards
Upstream configs are @strict dataclasses. Checkpoint keys upstream does not
declare survive as plain attributes, but only when the checkpoint sets them.
Defaults that only vLLM sets, and values vLLM derives, vanish.
| Hazard | Signal | Fix |
|---|---|---|
| Upstream consumes legacy keys | __post_init__ pops keys the model reads (deepseek_v4 compress_ratios) |
Read the upstream replacement (layer_types, mlp_layer_types, per-layer rope_parameters) |
| Renamed or dropped fields | Diff shows <MISSING> for a field the model reads |
Read the upstream name; derive dropped flags from what implies them (enable_moe_block from num_experts) |
| Undeclared field without a default | config.x raises AttributeError for configs that omit x, e.g. minimal test configs (A.X-K1 scoring_func). Dumps of real checkpoints miss it |
Upstream ignoring x usually means vLLM can stop reading it; if vLLM truly needs it, read getattr(config, "x", <old default>) |
| Top-level config stops forwarding | The vendored top-level config forwarded reads to text_config; upstream does not. getattr(config, "x", default) silently takes the default, e.g. on MTP draft configs from SpeculativeConfig.hf_config_override, which the dumps do not build |
Read from get_text_config(); grep getattr reads of every field, including speculative-decoding paths |
| Type check on the config | ctx.get_hf_config(SomeConfig) relies on the vendored base class |
Accept the upstream class too, e.g. a Cosmos3ProcessingInfo overriding get_hf_config |
| Remote code takes over | Checkpoint config.json has auto_map; with --trust-remote-code, the hub class replaces the native one once vLLM stops registering its own |
Check the remote class loads; if not, register the upstream class for that model_type |
| Read-only upstream fields | Properties or derived fields break subclasses and hf_overrides (nemotron_h hybrid_override_pattern) |
Override the source field (layers_block_type) instead |
| Behavioural rewrite | Upstream derives fields differently (diffusion_gemma sliding_window) |
Upstream matches the reference implementation; verify with an eval |
| Original checkpoint format | Upstream targets -hf conversions; vLLM serves the originals (qwen3_asr, qianfan_ocr) |
Keep only if upstream cannot load the original at all |
| Processor output contract | Model code consumes keys upstream does not emit | Port the model to upstream's keys |
| Name collision | A registered vendored processor shares an upstream class name and hijacks get_processor (InternVLProcessor) |
Drop the registry entry |
Version gates from the inventory need a decision:
-
Code only reachable below the floor is deleted. This covers:
- comparisons against
TRANSFORMERS_VERSION; - branches labelled
# Transformers < X; - fallbacks for fields that upstream configs now always set, e.g.
mlp_layer_types,per_layer_configor legacylayer_typesnames.
A
skipifwhose condition is now always true becomes an unconditionalskipwith the same reason. - comparisons against
-
min_transformers_versionat or below the floor is removed. -
max_transformers_versionbelow the floor means CI never tests that model; report it rather than deleting it silently. -
Patches citing a Transformers issue or PR:
- Delete the patch when its fix is in the floor release.
- For an issue, find the PR that closed it and check that PR. A closed issue is not a fixed one: issues get closed while the bug is still in the release.
- Re-run the patched behaviour before deleting. A fix in the release does not always cover vLLM's path: Emu3's processor gained BOS upstream, but vLLM tokenizes without the processor.
- Links that only explain a design choice (e.g. a float32 RoPE note) are not patches; leave them.
Testing
.venv/bin/python -m pytest tests/transformers_utils tests/config tests/tokenizers_
.venv/bin/python -m pytest tests/models/multimodal/processing/test_common.py -k "<models>"
Add tests/models/test_initialization.py -k "<archs>" where a GPU is
available, and a model eval for any model whose resolved config changed.
Models in vllm/models/ are high priority: whenever their config or model
code changes, run GSM8k (tests/evals/gsm8k/) and GPQA Diamond
(tests/evals/gpt_oss/) on main and the branch, even when the dumps show
no diff, and put both scores in the PR. FlashInfer autotuning picks kernel
tactics afresh on every server start, which moved DeepSeek-V4-Flash GSM8k by
about 1 point; pass --kernel-config '{"enable_flashinfer_autotune": false}'
to both servers so greedy outputs are comparable.