Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,53 @@
+---
+language:
+  - en
+license: apache-2.0
+base_model: unsloth/gpt-oss-20b-unsloth-bnb-4bit
+tags:
+  - fine-tuned
+  - constraint-extraction
+  - scheduling
+  - peft
+  - lora
+pipeline_tag: text-generation
+---
+
+# aischeduler-llm-20260607
+
+Fine-tuned unsloth/gpt-oss-20b-unsloth-bnb-4bit for structured employee scheduling constraint extraction.
+
+## Model description
+
+Extracts `softConstraints` JSON from free-text employee crew notes and manager limitation instructions.
+Output schema includes: `dailyTimeRestrictions`, `weeklyFrequencyLimits`, `consecutiveShiftLimits`,
+`recurringTimeOffPatterns`, `crossDayDependencies`, `advanceNoticeRequired`, `crewSizeRestrictions`,
+`leadershipRestrictions`, `jobTypeRestrictions`, `clientScheduleRestrictions`, `vehicleRestrictions`,
+`interpersonalConflicts`.
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch, json
+
+model = AutoModelForCausalLM.from_pretrained("loitranyuki/aischeduler-llm-20260607", torch_dtype=torch.float16, device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained("loitranyuki/aischeduler-llm-20260607")
+
+system_prompt = "..."  # see prompts/employee_constraint_extraction.txt
+user_text = "No evenings. Max 3 doubles per week."
+
+messages = [
+    {"role": "system", "content": system_prompt},
+    {"role": "user", "content": user_text},
+]
+tokens = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
+out = model.generate(tokens, max_new_tokens=512, temperature=0.1)
+raw = tokenizer.decode(out[0][tokens.shape[-1]:], skip_special_tokens=True)
+constraints = json.loads(raw)
+```
+
+## Training
+
+- **Framework:** Unsloth + PEFT (LoRA rank=16, alpha=32)
+- **Quantization:** 4-bit QLoRA during training, merged to bfloat16
+- **Target modules:** q_proj, k_proj, v_proj, o_proj