SourceHat RLCD
SourceHat RLCD returns typed categorical, binary, and ordinal decisions from
a shared Clef/Qwen-derived backbone. This edition contains the complete backbone
with all 496 trained text updates mathematically folded into its dense weights,
plus the FP32 joint decision head, in one model.safetensors file.
There are no adapter files, runtime LoRA branches, PEFT dependency, or separate base-model downloads. Tokenizer, configuration, and the inference wrapper are included.
Dense BF16 Deployment Check
Private review draft, not an all-gates-passed or publication-ready declaration.
These measurements describe the genuinely merged BF16 weights in this edition, not the earlier runtime-LoRA composition. Bold values are best within each matched population; ties are bold for both models.
| Population / model | Correct fields | Correct objects | Response coverage | Field NLL | Field Brier |
|---|---|---|---|---|---|
| Historical: unmerged reference | 410/439 | 230/256 | 100% | 0.151181 | 0.082997 |
| Historical: dense BF16 | 409/439 | 229/256 | 100% | 0.149168 | 0.080889 |
| Fresh: unmerged reference | 305/326 | 111/128 | 100% | 0.145430 | 0.077869 |
| Fresh: dense BF16 | 306/326 | 112/128 | 100% | 0.142819 | 0.076039 |
Both models have 100% response and probability-scoring coverage on both populations. These are bounded release checks, not the campaign's sealed final evaluation or a production-domain guarantee. The reference is the trained, unmerged model, not original Clef.
The separate numerical-fidelity check does not pass: four reload probes preserve all top-1 choices, but their maximum candidate-probability change is 0.055960, against the unchanged 0.020000 tolerance. Passing population-quality checks does not override this check. Over the complete historical and fresh populations, maximum probability changes are 0.106855 and 0.090278. The smaller four-probe value is not the global maximum.
Candidate-order and opaque-ID checks each cover 32 paired cases, with maximum logit differences 0 and 0. This tests unchanged descriptions and semantics, not arbitrary paraphrases.
Peak PyTorch-reserved memory is 53.13 GiB on one 64-GiB NVIDIA CMP 170HX. Mean warm request time is 2.551 seconds across 144 newly executed requests; cached predictions and model loading are excluded. This is not a validated single-16-GiB deployment.
Historical Training-Checkpoint Results
All models below scored the same 256 cases and 439 fields with 100% response coverage. These are historical training-checkpoint results. The dense edition's separate measurements and unresolved numerical-fidelity check are reported above.
| Model | Correct fields | Correct complete objects | Field NLL | Field Brier |
|---|---|---|---|---|
| Original full Clef | 395/439 (90.0%) | 216/256 (84.4%) | 0.2519 | 0.1268 |
| Trained head only | 403/439 (91.8%) | 223/256 (87.1%) | 0.2041 | 0.1070 |
| RLCD trained checkpoint | 410/439 (93.4%) | 230/256 (89.8%) | 0.1512 | 0.0830 |
The population contains 160 executable synthetic cases, 48 Banking77 cases, and 48 CLINC-OOS cases with restricted candidate lists. Training uses seed 126. The aggregate supplement distinguishes historical measurements from the current package's verification status.
Contract and Runtime
The decision contract accepts 1-4 text-only fields: categorical choices with 2-16 candidates, binary propositions with true then false, and ordered scales with 2-14 levels. The compiled limit is 2048 tokens; oversized requests are rejected, not truncated. Caller IDs are not model features.
The model proposes supplied choices and raw field distributions. It does not execute actions or independently estimate object correctness, safety, or workflow success. Supply applicable policies and meaningful fallback choices.
hf download SourceHat-Labs/RLCD-27B --local-dir ./sourcehat-rlcd
cd sourcehat-rlcd
python -m pip install -r requirements.txt
CUDA_VISIBLE_DEVICES=2 python inference.py --request example_request.json --device cuda:0
Select the physical GPU appropriate to your machine. Use the included typed-head loader, not a generic chat-generation loader. BF16 backbone storage and the FP32 head total 55.23 GB. The measured dense runtime and quality checks are reported above, including the unresolved numerical-fidelity check; this is not a single-16-GiB runtime claim.
License, Provenance, and Scope
The exact base/tokenizer is Cloudflare/clef revision
0b331204bb13fbd2ca93a64df1956f5b55478ce5. Training used supervised CE, Brier,
ordinal RPS, and a reference KL term, not policy-gradient RL. The package is
Apache-2.0; see LICENSE and NOTICE for attribution.
Training-source references: Banking77,
CC BY 4.0; CLINC-OOS, CC BY 3.0.
No training records, scripts, private prompts, logs, or intermediate checkpoints
are included. This is a research model, not a safety or production-risk certificate.
Packaging Identity
Version 1.0.0 uses the public typed-decision-v1 contract. CPU checks prove that the renamed wrapper preserves native compilation and fixed-logit responses. The BF16 and INT8 containers have branding-only header changes; tensor bytes are unchanged. EXL3 weights are byte-identical. Historical source hashes identify the original measured containers, not a newly executed quantization. The earlier offline CLI checks exercised the original wrappers; no new neural execution or evaluation was performed for this packaging migration. All model repositories and website candidates remain private/review-only.
See the SourceHat RLCD article.
- Downloads last month
- 20