pico-decoder-tiny-1 trained to 100k steps

README.md

@@ -0,0 +1,57 @@
+---
+datasets:
+- pico-lm/pretokenized-dolma
+language:
+- en
+license: apache-2.0
+metrics:
+- pico-lm/perplexity
+pipeline_tag: text-generation
+---
+
+# Pico Decoder Tiny
+
+**pico-decoder-tiny** is the smallest (11M) model in the `pico-decoder` suite — a lightweight, LLaMA-style decoder-only transformer trained from scratch using [`pico-train`](https://github.com/pico-lm/pico-train). It is designed for transparent and reproducible research into the learning dynamics of language models, and is fully compatible with the `pico-analyze` toolkit for detailed interpretability analysis.
+
+> NOTE: The `pico-decoder-tiny-1` branch contains the full commit history for the training run. 
+
+## 🔧 Model Details
+
+| Field               | Value                              |
+|---------------------|------------------------------------|
+| **Architecture**     | Decoder-only transformer (LLaMA-style) |
+| **Parameters**       | 11M                               |
+| **Layers**           | 12                                |
+| **Hidden Size**      | 96                               |
+| **Feed Foward Size** | 384                                |
+| **Attention Heads**  | 12                                 |
+| **Key/Value Heads**  | 4                                  |
+
+## 📚 Training
+
+- **Dataset**: [`pretokenized-dolma`](https://huggingface.co/datasets/pico-lm/pretokenized-dolma), English-only
+- **Training steps**: 200,000
+- **Batch size**: 1024
+- **Sequence length**: 2048
+- **Optimizer**: AdamW
+- **Learning rate schedule**: Linear decay with warmup
+- **Compute**: 16 A100-SXM4-80GB GPUs
+
+## 📈 Evaluation and Analysis
+
+This model supports fine-grained analysis using [`pico-analyze`](https://github.com/pico-lm/pico-analyze). This tool enables researchers to understand how learning unfolds over training, even at very small scales.
+
+We also evaluate perplexity of the model on the [`pico-paloma-tinsy`](https://huggingface.co/datasets/pico-lm/pretokenized-paloma-tinsy) dataset.
+
+## 📄 Citation
+
+If you use `pico-tiny` or any other `pico-decoder` model in your research, please cite:
+
+```bibtex
+@software{pico2025,
+    author = {Diehl Martinez, Richard},
+    title = {Pico: A Lightweight Framework for Studying Language Model Learning Dynamics},
+    year = {2025,
+    url = {https://github.com/pico-lm}
+}
+```

config.json

@@ -0,0 +1,22 @@
+{
+  "activation_hidden_dim": 384,
+  "architectures": [
+    "PicoDecoderHF"
+  ],
+  "attention_n_heads": 12,
+  "attention_n_kv_heads": 4,
+  "auto_map": {
+    "AutoConfig": "pico_decoder.PicoDecoderHFConfig",
+    "AutoModelForCausalLM": "pico_decoder.PicoDecoderHF"
+  },
+  "batch_size": 1024,
+  "d_model": 96,
+  "max_seq_len": 2048,
+  "model_type": "pico_decoder",
+  "n_layers": 12,
+  "norm_eps": 1e-06,
+  "position_emb_theta": 10000.0,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.3",
+  "vocab_size": 50304
+}

eval_results/step_0.json

@@ -0,0 +1 @@
+{"paloma": 59435.05139917247}

eval_results/step_1000.json

@@ -0,0 +1 @@
+{"paloma": 2176.422658291594}

eval_results/step_10000.json

@@ -0,0 +1 @@
+{"paloma": 313.8343901737226}

eval_results/step_100000.json

@@ -0,0 +1 @@
+{"paloma": 142.78638106821307}

eval_results/step_11000.json

@@ -0,0 +1 @@
+{"paloma": 268.332413412885}

eval_results/step_12000.json

@@ -0,0 +1 @@
+{"paloma": 254.46162965488767}

eval_results/step_13000.json

@@ -0,0 +1 @@
+{"paloma": 249.49274518747362}

eval_results/step_14000.json

@@ -0,0 +1 @@
+{"paloma": 260.20426084006704}

eval_results/step_15000.json

@@ -0,0 +1 @@
+{"paloma": 239.41960436525244}

eval_results/step_16000.json

@@ -0,0 +1 @@
+{"paloma": 229.52608692687562}

eval_results/step_17000.json

@@ -0,0 +1 @@
+{"paloma": 225.62338353731906}

eval_results/step_18000.json

@@ -0,0 +1 @@
+{"paloma": 212.63083450470236}

eval_results/step_19000.json

@@ -0,0 +1 @@
+{"paloma": 208.58393890899234}

eval_results/step_2000.json

@@ -0,0 +1 @@
+{"paloma": 884.8345587062504}

eval_results/step_20000.json

@@ -0,0 +1 @@
+{"paloma": 205.5417249480191}

eval_results/step_21000.json

@@ -0,0 +1 @@
+{"paloma": 201.91932611332538}

eval_results/step_22000.json

@@ -0,0 +1 @@
+{"paloma": 199.08593051392026}

eval_results/step_23000.json

@@ -0,0 +1 @@
+{"paloma": 198.37421456945066}

eval_results/step_24000.json

@@ -0,0 +1 @@
+{"paloma": 193.6218826051373}

eval_results/step_25000.json

@@ -0,0 +1 @@
+{"paloma": 192.3463352126942}

eval_results/step_26000.json

@@ -0,0 +1 @@
+{"paloma": 190.8057739201323}

eval_results/step_27000.json

@@ -0,0 +1 @@
+{"paloma": 192.40458190831572}

eval_results/step_28000.json

@@ -0,0 +1 @@
+{"paloma": 188.59788890599373}

eval_results/step_29000.json

@@ -0,0 +1 @@
+{"paloma": 187.05302696460632}

eval_results/step_3000.json

@@ -0,0 +1 @@
+{"paloma": 623.0875143150835}

eval_results/step_30000.json

@@ -0,0 +1 @@
+{"paloma": 186.20094004940074}

eval_results/step_31000.json

@@ -0,0 +1 @@
+{"paloma": 183.60527752766626}

eval_results/step_32000.json

@@ -0,0 +1 @@
+{"paloma": 182.63232411407841}

eval_results/step_33000.json

@@ -0,0 +1 @@
+{"paloma": 182.5158897050997}

eval_results/step_34000.json

@@ -0,0 +1 @@
+{"paloma": 562.858497242479}

eval_results/step_35000.json

@@ -0,0 +1 @@
+{"paloma": 184.30060736659513}

eval_results/step_36000.json

@@ -0,0 +1 @@
+{"paloma": 177.34308501486163}

eval_results/step_37000.json

@@ -0,0 +1 @@
+{"paloma": 177.5313984960629}

eval_results/step_38000.json

@@ -0,0 +1 @@
+{"paloma": 178.25564483549536}

eval_results/step_39000.json

@@ -0,0 +1 @@
+{"paloma": 174.9592615609385}

eval_results/step_4000.json

@@ -0,0 +1 @@
+{"paloma": 606.9779211090832}

eval_results/step_40000.json

@@ -0,0 +1 @@
+{"paloma": 174.86358099016994}

eval_results/step_41000.json

@@ -0,0 +1 @@
+{"paloma": 173.4716007269218}

eval_results/step_42000.json

@@ -0,0 +1 @@
+{"paloma": 174.3865502098296}

eval_results/step_43000.json

@@ -0,0 +1 @@
+{"paloma": 171.96283629081805}

eval_results/step_44000.json

@@ -0,0 +1 @@
+{"paloma": 172.43479460373987}

eval_results/step_45000.json

@@ -0,0 +1 @@
+{"paloma": 170.67068021986958}

eval_results/step_46000.json

@@ -0,0 +1 @@
+{"paloma": 170.5537706348539}

eval_results/step_47000.json

@@ -0,0 +1 @@
+{"paloma": 168.9578624722019}

eval_results/step_48000.json

@@ -0,0 +1 @@
+{"paloma": 169.10025308456156}

eval_results/step_49000.json

@@ -0,0 +1 @@
+{"paloma": 168.10817976263758}

eval_results/step_5000.json

@@ -0,0 +1 @@
+{"paloma": 1044.7105118223183}

eval_results/step_50000.json

@@ -0,0 +1 @@
+{"paloma": 167.72806381398377}