svyatsharov
/

Role-play-ai

@@ -1,210 +1,204 @@
 ---
-base_model: unsloth/qwen2.5-32b-instruct-bnb-4bit
 library_name: peft
 pipeline_tag: text-generation
-tags:
-- base_model:adapter:unsloth/qwen2.5-32b-instruct-bnb-4bit
-- lora
-- sft
-- transformers
-- trl
-- unsloth
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]
-### Framework versions
-- PEFT 0.19.1

 ---
+license: apache-2.0
+base_model: Qwen/Qwen2.5-32B-Instruct
+tags:
+  - lora
+  - peft
+  - sft
+  - trl
+  - unsloth
+  - roleplay
+  - conversational
+  - russian
+  - english
+language:
+  - en
+  - ru
 library_name: peft
 pipeline_tag: text-generation
+datasets:
+  - PygmalionAI/PIPPA
+  - IlyaGusev/saiga_scored
+  - Norquinal/claude_multiround_chat_30k
 ---
+# Role-play-ai
+Fine-tuned `Qwen/Qwen2.5-32B-Instruct` для SFW roleplay-диалогов на русском и английском. Адаптер LoRA r=64.
+**Code & metrics**: https://github.com/ichinosekei/Role-play-ai
+## Quick start
+### PEFT
+```python
+from peft import PeftModel
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+base = AutoModelForCausalLM.from_pretrained(
+    "Qwen/Qwen2.5-32B-Instruct",
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+)
+model = PeftModel.from_pretrained(base, "svyatsharov/Role-play-ai")
+tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-32B-Instruct")
+messages = [
+    {"role": "system", "content": "You are Mira, a warm tavern owner. Witty but firm."},
+    {"role": "user", "content": "*sits at the bar* Tough day."},
+]
+inputs = tokenizer.apply_chat_template(messages, tokenize=True,
+                                       add_generation_prompt=True,
+                                       return_tensors="pt").to(model.device)
+out = model.generate(inputs, max_new_tokens=300, temperature=0.85, top_p=0.9, do_sample=True)
+print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))
+```
+### Unsloth (быстрее на 1 GPU)
+```python
+from unsloth import FastLanguageModel
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name="svyatsharov/Role-play-ai",
+    max_seq_length=4096,
+    load_in_4bit=True,
+)
+FastLanguageModel.for_inference(model)
+```
+## Model details
+| Параметр | Значение |
+|---|---|
+| Base model | `Qwen/Qwen2.5-32B-Instruct` |
+| Adapter type | LoRA (PEFT) r=64, alpha=128 |
+| Trainable params | 537M (1.6% от 32B) |
+| Context length | 6144 |
+| Languages | English, Russian |
+| Chat template | ChatML |
+| License | Apache 2.0 (наследуется от Qwen2.5) |
+## Evaluation results
+Метрики посчитаны на eval-сете (1203 примера, 5% от полного датасета). Все 5 групп метрик:
+### Loss-based
+| Метрика | Значение | Цель |
+|---|---|---|
+| Perplexity (overall) | 3.31 | 5–12 |
+| Perplexity (EN) | 3.24 | 5–12 |
+| Perplexity (RU) | 3.58 | 5–15 |
+| Token accuracy | 0.678 | >0.55 ✅ |
+### Reference-based
+| Метрика | Значение | Цель |
+|---|---|---|
+| BLEU-4 | 11.20 | >5 ✅ |
+| ROUGE-L | 0.215 | >0.20 ✅ |
+| BERTScore F1 | **0.865** | >0.85 ✅ |
+| chrF++ | 31.01 | >25 ✅ |
+### Style match (Albert metrics)
+| Метрика | Значение | Цель |
+|---|---|---|
+| Length JS-divergence | 0.038 | <0.10 ✅ |
+| Vocabulary overlap | 0.94 | >0.55 ✅ |
+| **Style Match Score** | **0.798** | >0.7 ✅ |
+### Diversity
+| Метрика | Значение | Цель |
+|---|---|---|
+| Avg distinct-2 | 0.940 | >0.7 ✅ |
+| Avg distinct-3 | 0.985 | — |
+| Self-repetition rate | 0.000 | <0.05 ✅ |
+| TTR (model) | 0.349 | 0.4–0.6 |
+### Caveats
+- **PPL 3.31 ниже целевых 5–12** — eval-сет это случайные 5% из тех же источников что и train. На out-of-distribution данных PPL будет выше. Не настоящее переобучение: train_loss = 1.20, eval_loss = 1.21.
+- **Русский в eval недопредставлен** — всего 14 RU-примеров из 1203. Метрика PPL_ru статистически слабая.
+## Training data
+Целевой объём: 50k диалогов SFW roleplay. Реально получено **24 071**.
+| Источник | Получено | % | Заявлено |
+|---|---|---|---|
+| [`PygmalionAI/PIPPA`](https://huggingface.co/datasets/PygmalionAI/PIPPA) (SFW filter) | 13 050 | 54.2% | 35% |
+| `lemonilia/LimaRP` | 0 | 0% | 25% |
+| [`Norquinal/claude_multiround_chat_30k`](https://huggingface.co/datasets/Norquinal/claude_multiround_chat_30k) | 7 500 | 31.2% | 15% |
+| [`IlyaGusev/saiga_scored`](https://huggingface.co/datasets/IlyaGusev/saiga_scored) | 3 521 | 14.6% | 25% |
+**LimaRP не загрузился** при подготовке датасета — fallback увеличил долю PIPPA.
+## Training procedure
+| Параметр | Значение |
+|---|---|
+| Framework | Unsloth + TRL SFTTrainer |
+| Method | QLoRA 4-bit (NF4) |
+| Effective batch | 16 (2 × 8 grad_accum) |
+| Epochs | 3 (early stopping) |
+| Learning rate | 1e-4, cosine schedule, warmup 3% |
+| Optimizer | AdamW 8-bit |
+| Weight decay | 0.01 |
+| Gradient checkpointing | Unsloth |
+| Precision | bf16 + tf32 |
+| Target modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
+### Steps & duration
+- Total steps: 4287
+- Train duration: **51 час** на 1× A100 80GB
+- Peak VRAM: 33 ГБ
+- Final train loss: 1.20
+- Final eval loss: 1.21
+## Hardware
+- **GPU**: 1× NVIDIA A100 80GB
+- **RAM**: 200 GB
+- **Storage**: 4 TB
+- **VRAM лимит**: 60 GB (использовано 33 GB)
+## Limitations
+- Английский лучше русского из-за дисбаланса данных
+- 32B модель требует ≥24 GB VRAM для инференса в 4-bit
+- Контекст 6144 токенов — длинные RP-сессии надо обрезать
+- Eval-сет близок к train → реальное качество на новых данных вероятно ниже метрик
+- LimaRP не вошёл в обучение — состав датасета смещён
+## Citation
+```bibtex
+@misc{role-play-ai-2026,
+  title  = {Role-play AI: Qwen2.5-32B fine-tune for bilingual SFW roleplay},
+  author = {svyatsharov, ichinosekei},
+  year   = {2026},
+  url    = {https://huggingface.co/svyatsharov/Role-play-ai}
+}
+```
+Base model citation:
+```bibtex
+@misc{qwen2.5,
+  title  = {Qwen2.5: A Party of Foundation Models},
+  author = {Qwen Team},
+  year   = {2024},
+  url    = {https://huggingface.co/Qwen/Qwen2.5-32B-Instruct}
+}
+```
+## Framework versions
+- PEFT 0.19.1
+- Transformers ≥4.46.0
+- TRL ≥0.12.0
+- Unsloth
+- PyTorch 2.5.1