Model card with training details and dataset provenance

README.md

@@ -25,7 +25,7 @@ model-index:
 
 **Dr. Sage** is a fine-tuned version of [Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct) trained to act as a probing, honest, and empathetic therapeutic companion.
 
-Trained: **2026-03-19** | Base: `Qwen2.5-3B-Instruct` | Loss: `0.5166`
+Trained: **2026-03-19** | Base: `Qwen2.5-3B-Instruct` | Loss: `0.0000`
 
 ---
 
@@ -50,30 +50,13 @@ Dr. Sage never lectures. Never gives long speeches. Never asks more than **one q
 
 | Category | Samples |
 |----------|---------|
-| relationships | 675 |
-| general_therapeutic | 596 |
-| depression | 558 |
-| anxiety | 544 |
-| family_dynamics | 414 |
-| trauma | 333 |
-| work_burnout | 251 |
-| anger | 250 |
-| loneliness | 243 |
-| substance_avoidance | 242 |
-| shame_esteem | 232 |
-| grief_loss | 230 |
-| identity_transition | 229 |
-| self_esteem | 225 |
-| boundaries_pleasing | 225 |
-| crisis_safety | 30 |
-| physical_somatic | 10 |
+| general_therapeutic | 5,287 |
 
 ### By source
 
 | Source | Samples |
 |--------|---------|
-| synthetic | 4,375 |
-| alpaca | 912 |
+| synthetic + alpaca | 5,287 |
 
 Full dataset available at [Phora68/dr-sage-dataset](https://huggingface.co/datasets/Phora68/dr-sage-dataset)
 
@@ -95,8 +78,8 @@ Full dataset available at [Phora68/dr-sage-dataset](https://huggingface.co/datas
 | LR schedule | cosine |
 | Optimizer | adamw_8bit |
 | Hardware | A100 80GB |
-| Final loss | 0.5166 |
-| Training time | 15 min |
+| Final loss | 0.0000 |
+| Training time | 0 min |
 
 ---