znzryb/tie-qwen2.5-1.5b-length-predictor
text-classification model by znzryb
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
BERT-base uncased with two raw output logits: mu = logits[:,0], sigma = softplus(logits[:,1]) + 0.05. Fixed Student-t degrees of freedom 3.5. The logits are distribution parameters, NOT classification probabilities or token lengths. Load with AutoModelForSequenceClassification and AutoTokenizer. Set tokenizer.truncationside = "left", use maxlength=512, eval() and no gradients. Use distribution.py scores(mu, sigma, beta=0.1, alpha=0.9, cap=8192). Bounded TIE = E[min(max(exp(Y),1),8192)] + beta CVaR0.9 of that served length. An unbounded log-t has no finite arithmetic expectation. Score integration is deterministic. Target: Qwen/Qwen2.5-1.5B-Instruct at revision 989aa7980e4cf806f80c7fef2b1adb7bc71aa306. 8,000 training, 1,000 validation, 1,000 internal and 1,000 external test prompts. Three training epochs, censored log-t NLL; first epoch selected on validation NLL. Positive beta selected on validation ranking; beta=0 is reported as the mean-only ablation. No Qwen generation was run for this experiment: all full-v1 greedy labels were reused. This is NOT the paper author's repeated-stochastic-sampling parameter-label recipe.…
Read the full model card ↗ · Preview checked 2026-10-27T23:36:06.196Z
Inside the original model card — Document outline
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Links from the model card
References supplied by the publisher, not independently verified endorsements. Check the destination before downloading files or entering credentials.
Documentation belongs to its respective authors. Reported project/model license: apache-2.0. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
znzryb/tie-qwen2.5-1.5b-length-predictor is a text-classification model repository published by znzryb on Hugging Face. The source reports the transformers library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- text-classification
- Library
- transformers
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- 109,483,778
- Architecture
- BertForSequenceClassification
- License
- apache-2.0
- Access
- Not gated by Hugging Face
- Created
- 2026-10-15T06:38:32.000Z
- Last modified
- 2026-10-15T06:42:05.000Z
Compatibility and lineage
Reported languages: en
Reported base models: google-bert/bert-base-uncased
Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.
Use and evaluate
Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.
No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-10-15T06:42:18.893Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.