Xnhyacinth/LongWorld-Checkpoints

LongWorld Checkpoints

Three final step-680 related-work baseline exports and three LongWorld research-selected exports. Each subfolder contains its own model shards, index, configuration, tokenizer and processor files. The variants retain independent learned weights.

Subfolder	Starting model	SFT data
acc-step680	Qwen3.5-4B Instruct	ACC
base-acc-step680	Qwen3.5-4B-Base	ACC
base-longtrace-step680	Qwen3.5-4B-Base	LongTraceRL

These are the final aligned step-680 exports, not a selection of the best validation checkpoint. The Base LongTrace run was trained independently from the Base ACC run. The training inputs are in the 128k baseline dataset. They are related-work baselines, not checkpoints trained on the LongWorld workflow-probe product.

Download one variant

from huggingface_hub import snapshot_download

root = snapshot_download(
    "Xnhyacinth/LongWorld-Checkpoints",
    allow_patterns=["base-acc-step680/*"],
    local_dir="longworld-checkpoints",
)

Use the selected subfolder with a Qwen3.5-compatible inference stack. Structural validation covers the complete safetensors headers, shard index and loading files; inference has not been executed as part of this publication review. ARTIFACTS.json records the exact published paths, sizes and SHA-256 identities.

Each baseline subfolder carries the weight and loading files plus its training recipe (see Standalone model cards below). Numeric training metrics are retained in a private research record. Optimizer/RNG state, historical working snapshots and the removed training_args.bin pickles are excluded. The original file hashes are retained for the selected exports. Shared tokenizer files appear in each subfolder so each variant can be downloaded independently; no storage-saving claim is made.

The source exports and the upstream Qwen3.5-4B and Qwen3.5-4B-Base cards declare Apache-2.0. The companion training datasets retain their own terms.

Standalone model cards

Each subfolder is a complete, independently loadable model card: it carries the variant's own args.json (full training recipe), CLEANUP_20260930_RESTORE_MAP.json (restore map for the removed training_args.bin pickle) and a README.md with the run summary. Per-step numeric training and evaluation histories are retained in private research records; current model folders no longer contain trainer_state.json. Loading a subfolder needs no other repository.

from transformers import AutoProcessor, AutoModelForCausalLM

repo = "Xnhyacinth/LongWorld-Checkpoints"
subfolder = "base-acc-step680"  # or "acc-step680" / "base-longtrace-step680"
processor = AutoProcessor.from_pretrained(repo, subfolder=subfolder)
model = AutoModelForCausalLM.from_pretrained(
    repo, subfolder=subfolder, dtype="auto", device_map="auto"
)

All three runs trained for 680 steps at max_length 133120. The two ACC runs report both train and eval loss; the LongTrace run reports eval loss only. Train loss is the mean over the 137 logged steps unless stated otherwise.

Variant	Starting model	Data	Train loss	Eval loss (step 680)
acc-step680	Qwen3.5-4B (instruct)	ACC	0.3645 (final step; run mean 0.4327)	0.37456
base-acc-step680	Qwen3.5-4B-Base	ACC	0.4382 (run mean; final step 0.3725)	0.37545
base-longtrace-step680	Qwen3.5-4B-Base	LongTraceRL	0.0623 (run mean; final step 0.0040)	0.17741

Notes carried over from the standalone cards:

• acc-step680 — GBS 16, 8 GPU (SP=4, DP=2), lr 1e-5 cosine, max_length 133120, 680 steps; source data/sft/swift_ext_acc/v6-20260825-153450.
• base-acc-step680 — GBS 16, SP 4, micro 1, accum 16, lr 1e-5 cosine, max_length 133120, 680 steps; source data/sft/swift_ext_acc_base/v1-20260903-163953.
• base-longtrace-step680 — same recipe as Base ACC; trained independently of ACC (not continued from it); resumed from checkpoint-300 after a 24h session abort. Best eval during the run was checkpoint-200 (0.129); this subfolder is the aligned step-680 export, not the best-validation checkpoint.

All three omit optimizer/RNG state and the training_args.bin pickle.

LongWorld research-selected inference exports

Three independent inference-only exports are available in the research-selected/ folder:

Subfolder	Role	Evidence boundary
p456-base-step1366	Completed P456 final endpoint	The recorded four-model evaluation used unequal training budgets and does not establish causal data efficacy.
p64-base-step200	P64 trainer-selected best validation loss	Selected by training validation loss, not included in the cited completed step-680 downstream run.
p64-base-step680	P64 completed final endpoint	Historical GraphWalks precision/recall and parse-zero F1 were zero with 100% format failure.

Each subfolder includes model shards, index, configuration, tokenizer and processor files; no optimizer, scheduler, RNG or resume shards. The exact file SHA-256 values and source binding are in the selected export manifest. Tensor headers and indices were checked; no new GPU inference was performed during this publication. P64's original input manifest classifies the source as local-probe and does not establish production eligibility or strict long-dependency verification. A scan of the P64 source data found security-related string hints, including names used in code examples; it did not establish whether those strings are valid credentials. Training-data source permissions have not been exhaustively resolved, and model memorization of sensitive material was not tested. The P456 and P64 exports are not interchangeable with the baseline variants above.

For an individual model, download only its subfolder, for example:

from huggingface_hub import snapshot_download

snapshot_download("Xnhyacinth/LongWorld-Checkpoints",
                  allow_patterns=["research-selected/p456-base-step1366/*"],
                  local_dir="/volume/pt-dev/qjiu/longworld-models")