Add canonical model card — v6 Kaggle T4, 93.95% accuracy

README.md

@@ -4,9 +4,9 @@ language:
 license: apache-2.0
 tags:
 - text-classification
-- emotion-classification
-- roberta
 - the-founder
+- emotion
+- 6-class
 datasets:
 - dair-ai/emotion
 metrics:
@@ -18,9 +18,9 @@ model-index:
   results:
   - task:
       type: text-classification
-      name: Text Classification
+      name: Emotion Classification
     dataset:
-      name: dair-ai/emotion
+      name: Emotion
       type: dair-ai/emotion
       split: validation
     metrics:
@@ -28,7 +28,7 @@ model-index:
       value: 0.9395
       name: Accuracy
     - type: loss
-      value: 0.2677
+      value: 0.2680
       name: Eval Loss
 ---
 
@@ -38,57 +38,69 @@ Fine-tuned by **The Founder** — an autonomous ML orchestration superagent.
 
 ## Model Description
 
-Fine-tuned `roberta-base` on `dair-ai/emotion` for 6-class emotion classification (sadness, joy, love, anger, fear, surprise).
+This is a fine-tuned version of `roberta-base` on the `dair-ai/emotion` dataset for Emotion Classification.
+
+Training was orchestrated end-to-end by **The Founder** — a personal ML agent that handles research, compute scheduling, experiment tracking, and artifact management autonomously using Kaggle (Tesla T4), Weights & Biases, and HuggingFace Hub.
 
 ## Model Details
 
 | Property | Value |
 |---|---|
 | Base model | roberta-base |
-| Dataset | dair-ai/emotion |
+| Fine-tuned on | dair-ai/emotion |
+| Task | Emotion Classification |
 | Epochs | 4 |
 | Batch size | 32 |
 | Learning rate | 2e-05 |
 | GPU | Tesla T4 (Kaggle) |
-| Train loss | 0.7915 |
-| Eval loss | 0.2677 |
+| Train loss | 0.2018 |
+| Eval loss | 0.2680 |
 | Eval accuracy | 0.9395 |
-| Duration | 12.3 min |
+| Duration | 21.3 min |
 
 ## How to Get Started
 
 ```python
 from transformers import pipeline
 clf = pipeline("text-classification", model="zanesmit29/founder-roberta-emotion-v1")
-clf("I am so happy today!")
+clf("Your input text here")
 ```
 
 ## Uses
 
-Emotion detection across sadness, joy, love, anger, fear, surprise.
+### Direct Use
+This model is suitable for Emotion Classification tasks in English. It can be used out-of-the-box with the Transformers pipeline API.
+
+### Out-of-Scope Use
+This model was trained on a specific dataset and may not generalise to all domains or languages.
+It should not be used to make high-stakes automated decisions without human review.
+Performance on out-of-distribution data (e.g. non-English text, domain-specific jargon) is not guaranteed.
 
 ## Training Details
 
 ### Data
-Trained on `dair-ai/emotion` — 4 epochs.
+Trained on `dair-ai/emotion` — 4 epochs, no additional preprocessing beyond standard tokenization.
 
 ### Hyperparameters
 
-| Param | Value |
+| Hyperparameter | Value |
 |---|---|
-| LR | 2e-05 |
-| Batch | 32 |
+| Learning rate | 2e-05 |
+| Batch size | 32 |
 | Epochs | 4 |
+| Optimizer | AdamW |
+| LR scheduler | Linear with warmup |
+| Max sequence length | 128 |
 | fp16 | true |
 
 ## Results
 
 | Metric | Value |
 |---|---|
-| Train loss | 0.7915 |
-| Eval loss | 0.2677 |
+| Train loss | 0.2018 |
+| Eval loss | 0.2680 |
 | Eval accuracy | 0.9395 |
-| Verdict | CONFIRMED |
+| Duration | 21.3 min |
 
 ## Experiment Tracking
 
@@ -98,7 +110,7 @@ Trained on `dair-ai/emotion` — 4 epochs.
 
 | Component | Tool |
 |---|---|
-| Compute | Tesla T4 (Kaggle) |
-| Tracking | Weights & Biases |
-| Artifacts | HuggingFace Hub |
+| Compute | Kaggle (Tesla T4) |
+| Experiment tracking | Weights & Biases |
+| Artifact storage | HuggingFace Hub |
 | Orchestration | The Founder |