Update README.md
Browse files
README.md
CHANGED
|
@@ -15,6 +15,8 @@ tags:
|
|
| 15 |
|
| 16 |
# SAT Tutor — Qwen3-8B adapter
|
| 17 |
|
|
|
|
|
|
|
| 18 |
A QLoRA fine-tune of `Qwen/Qwen3-8B` that tutors for the SAT (math + reading & writing).
|
| 19 |
Trained on ~17k SAT-style examples (60% math, 40% reading & writing), one epoch,
|
| 20 |
2048-token training length, with full step-by-step solutions plus hint-style answers.
|
|
@@ -73,3 +75,7 @@ For local/phone inference (llama.cpp, Ollama, LM Studio), use the GGUF build:
|
|
| 73 |
- SAT-focused: off-topic questions get base-model-quality answers at best.
|
| 74 |
- Reading & writing trails math — see the eval table above.
|
| 75 |
- It can still make mistakes — double-check the math against the steps shown.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
# SAT Tutor — Qwen3-8B adapter
|
| 17 |
|
| 18 |
+
by **AGK FIRE INC**
|
| 19 |
+
|
| 20 |
A QLoRA fine-tune of `Qwen/Qwen3-8B` that tutors for the SAT (math + reading & writing).
|
| 21 |
Trained on ~17k SAT-style examples (60% math, 40% reading & writing), one epoch,
|
| 22 |
2048-token training length, with full step-by-step solutions plus hint-style answers.
|
|
|
|
| 75 |
- SAT-focused: off-topic questions get base-model-quality answers at best.
|
| 76 |
- Reading & writing trails math — see the eval table above.
|
| 77 |
- It can still make mistakes — double-check the math against the steps shown.
|
| 78 |
+
|
| 79 |
+
---
|
| 80 |
+
|
| 81 |
+
© 2026 AGK FIRE INC. Released under Apache 2.0.
|