jeremierostan/Qwen3-1.7B-teacher-note-severity-GGUF

Qwen3-1.7B Teacher Note Severity — GGUF

GGUF builds of jeremierostan/Qwen3-1.7B-teacher-note-severity, a LoRA fine-tune of Qwen/Qwen3-1.7B (32K context) that classifies teacher notes about student behavior and academics.

Files:

• —Qwen3-1.7B-teacher-note-severity-f16.gguf (3.4 GB) — full precision
• —Qwen3-1.7B-teacher-note-severity-q8_0.gguf (1.8 GB) — recommended; near-identical quality, half the size

Output format

Given a teacher note (or a running log of notes), the model replies with a single JSON object:

{"category": "commendation | misbehavior | academic_concern", "severity": <int -100..100>, "escalate": <bool>}

Commendations score negative; routine notes 5–55; severity >= 60 means escalate to admin (the escalate flag is set exactly when severity >= 60, across all categories).

LM Studio

1. 1.In LM Studio, search this repo id (jeremierostan/Qwen3-1.7B-teacher-note-severity-GGUF) in the Hugging Face search tab and download the q8_0 file, or place the file in your models folder.
2. 2.Load the model; any context length from 2k up works for single notes — raise it (up to 32k) to feed long running logs of many notes.
3. 3.Set the system prompt (see below). In model settings, turn off thinking mode / keep the default Qwen3 non-thinking setting so the answer is pure JSON.
4. 4.Paste a note. Expected reply: JSON only.

System prompt

You are a school record assistant. Classify the teacher note about a student. Respond with ONLY a JSON object with keys: "category" (one of "commendation", "misbehavior", "academic_concern"), "severity" (integer, -100 to 100: commendations are negative, routine notes are 5-55, serious concerns warranting admin attention are 60 or above), and "escalate" (true only if severity >= 60).

Notes

• —Trained on synthetic data (jeremierostan/teacher-notes-severity); validate on real notes before production use.
• —Severity within a band is approximate (MAE ~17 on the held-out synthetic test; band accuracy 91%, escalation F1 1.0).