pico-decoder-medium-1 trained to 125k steps

README.md

@@ -0,0 +1,54 @@
+---
+datasets:
+- pico-lm/pretokenized-dolma
+language:
+- en
+license: apache-2.0
+metrics:
+- pico-lm/perplexity
+pipeline_tag: text-generation
+---
+
+# Pico Decoder Medium
+
+**pico-decoder-medium** is a 181M parameter model in the `pico-decoder` suite, balancing scale and analyzability. Built with [`pico-train`](https://github.com/pico-lm) and instrumented with [`pico-analyze`](https://github.com/pico-lm), it enables detailed studies of layer-wise learning behavior during language model pretraining.
+
+> NOTE: The `pico-decoder-medium-1` branch contains the full commit history for the training run. 
+
+## 🔧 Model Details
+
+| Field               | Value                              |
+|---------------------|------------------------------------|
+| **Architecture**     | Decoder-only transformer (LLaMA-style) |
+| **Parameters**       | 181M                              |
+| **Layers**           | 12                                |
+| **Hidden Size**      | 768                               |
+| **Feed Forward Size**| 3072                              |
+| **Attention Heads**  | 12                                 |
+| **Key/Value Heads**  | 4                                 |
+
+## 📚 Training
+
+- **Dataset**: [`pretokenized-dolma`](https://github.com/pico-lm)
+- **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). 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
+
+```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": 3072,
+  "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": 768,
+  "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": 59416.7212543554}

eval_results/step_1000.json

@@ -0,0 +1 @@
+{"paloma": 499.5738274564311}

eval_results/step_10000.json

@@ -0,0 +1 @@
+{"paloma": 51.102903796073036}

eval_results/step_100000.json

@@ -0,0 +1 @@
+{"paloma": 30.98631727670543}

eval_results/step_101000.json

@@ -0,0 +1 @@
+{"paloma": 30.9158632586642}

eval_results/step_102000.json

@@ -0,0 +1 @@
+{"paloma": 30.87562179640195}

eval_results/step_103000.json

@@ -0,0 +1 @@
+{"paloma": 30.830084008123816}

eval_results/step_104000.json

@@ -0,0 +1 @@
+{"paloma": 30.771775771350395}

eval_results/step_105000.json

@@ -0,0 +1 @@
+{"paloma": 30.77101141268368}

eval_results/step_106000.json

@@ -0,0 +1 @@
+{"paloma": 30.71518267995389}

eval_results/step_107000.json

@@ -0,0 +1 @@
+{"paloma": 30.651495790315423}

eval_results/step_108000.json

@@ -0,0 +1 @@
+{"paloma": 30.646850140418742}

eval_results/step_109000.json

@@ -0,0 +1 @@
+{"paloma": 30.60034494516326}

eval_results/step_11000.json

@@ -0,0 +1 @@
+{"paloma": 48.70718716470207}

eval_results/step_110000.json

@@ -0,0 +1 @@
+{"paloma": 30.57924894937655}

eval_results/step_111000.json

@@ -0,0 +1 @@
+{"paloma": 30.548662317006844}

eval_results/step_112000.json

@@ -0,0 +1 @@
+{"paloma": 30.50031009675734}

eval_results/step_113000.json

@@ -0,0 +1 @@
+{"paloma": 30.47728565412116}

eval_results/step_114000.json

@@ -0,0 +1 @@
+{"paloma": 30.407706258355116}

eval_results/step_115000.json

@@ -0,0 +1 @@
+{"paloma": 30.382002816715307}

eval_results/step_116000.json

@@ -0,0 +1 @@
+{"paloma": 30.30816983603434}

eval_results/step_117000.json

@@ -0,0 +1 @@
+{"paloma": 30.328557960330816}

eval_results/step_118000.json

@@ -0,0 +1 @@
+{"paloma": 30.25960363577467}

eval_results/step_119000.json

@@ -0,0 +1 @@
+{"paloma": 30.239006059759586}

eval_results/step_12000.json

@@ -0,0 +1 @@
+{"paloma": 48.04790251246728}

eval_results/step_120000.json

@@ -0,0 +1 @@
+{"paloma": 30.233270551435623}

eval_results/step_121000.json

@@ -0,0 +1 @@
+{"paloma": 30.18521380598952}

eval_results/step_122000.json

@@ -0,0 +1 @@
+{"paloma": 30.14745272227696}

eval_results/step_123000.json

@@ -0,0 +1 @@
+{"paloma": 30.138502722334778}

eval_results/step_124000.json

@@ -0,0 +1 @@
+{"paloma": 30.074931180435605}

eval_results/step_125000.json

@@ -0,0 +1 @@
+{"paloma": 30.083627031821408}

eval_results/step_13000.json

@@ -0,0 +1 @@
+{"paloma": 45.9789450450226}

eval_results/step_14000.json

@@ -0,0 +1 @@
+{"paloma": 45.154312149988236}

eval_results/step_15000.json

@@ -0,0 +1 @@
+{"paloma": 44.130179383032}

eval_results/step_16000.json

@@ -0,0 +1 @@
+{"paloma": 43.383745260105734}

eval_results/step_17000.json

@@ -0,0 +1 @@
+{"paloma": 42.70362300017154}

eval_results/step_18000.json

@@ -0,0 +1 @@
+{"paloma": 42.00629499373951}

eval_results/step_19000.json

@@ -0,0 +1 @@
+{"paloma": 41.85491885225117}

eval_results/step_2000.json

@@ -0,0 +1 @@
+{"paloma": 177.08616091166638}

eval_results/step_20000.json

@@ -0,0 +1 @@
+{"paloma": 40.92828142551595}

eval_results/step_21000.json

@@ -0,0 +1 @@
+{"paloma": 40.3970818064354}

eval_results/step_22000.json

@@ -0,0 +1 @@
+{"paloma": 40.0735236918054}

eval_results/step_23000.json

@@ -0,0 +1 @@
+{"paloma": 39.55490014910283}

eval_results/step_24000.json

@@ -0,0 +1 @@
+{"paloma": 39.20364000381908}

eval_results/step_25000.json

@@ -0,0 +1 @@
+{"paloma": 38.902612380283635}

eval_results/step_26000.json

@@ -0,0 +1 @@
+{"paloma": 38.46148998878559}

eval_results/step_27000.json

@@ -0,0 +1 @@
+{"paloma": 38.17645851178452}

eval_results/step_28000.json

@@ -0,0 +1 @@
+{"paloma": 37.92720969861393}