Add dataset card and loading script

README.md

@@ -0,0 +1,106 @@
+---
+language:
+- en
+license: apache-2.0
+task_categories:
+- other
+tags:
+- gemma
+- titans
+- activations
+- hidden-states
+- precomputed
+size_categories:
+- 100K<n<1M
+---
+
+# OpenWebText — Gemma-3-1B Hidden State Activations (Layer 23)
+
+Precomputed hidden state activations from layer 23 of [Gemma-3-1B-IT](https://huggingface.co/google/gemma-3-1b-it) for the [OpenWebText](https://huggingface.co/datasets/Skylion007/openwebtext) dataset, tokenized with sequence length 1024.
+
+Designed for training a **[Titans](https://arxiv.org/abs/2501.00663)** memory layer inserted after layer 23 of Gemma 3.
+
+## Dataset Structure
+
+Each shard contains pre-computed forward pass outputs up to layer 23:
+
+| File | Shape | Dtype | Description |
+|------|-------|-------|-------------|
+| `shard_NNNNNN.npy` | `(64, 1024, 3072)` | `bfloat16` | Hidden state activations |
+| `shard_NNNNNN_masks.npy` | `(64, 1024)` | `int32` | Attention masks (1=real, 0=pad) |
+| `shard_NNNNNN_tokens.npy` | `(64, 1024)` | `int32` | Token IDs |
+
+- **Shards:** 1121 (000000–001120)
+- **Examples per shard:** 64
+- **Total examples:** ~71,744
+- **Sequence length:** 1024
+- **Hidden dimension:** 3072 (Gemma-3-1B embed_dim)
+- **Source model:** `google/gemma-3-1b-it`
+- **Source dataset:** `veriga/openwebtext-gemma3-tokenized-1024`
+- **Total size:** ~153 GB
+
+## How It Was Created
+
+Activations were computed using a truncated forward pass through the first 23 Gemma 3 transformer layers. The process:
+
+1. Load OpenWebText tokens from `veriga/openwebtext-gemma3-tokenized-1024`
+2. Pad/truncate to 1024 tokens, generate attention masks
+3. Forward pass through layers 0–22 of Gemma-3-1B-IT
+4. Apply attention mask: `hidden * mask[:, :, None]` (zero out padding positions)
+5. Save as `.npy` shards in `bfloat16`
+
+See the [precomputation notebook](https://github.com/andrew-veriga/Titans_jax/blob/main/colabs/Precompute_Activations_Gemma3_GPU.ipynb) for full details.
+
+## Loading the Dataset
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("veriga/openwebtext-gemma3-tokenized-1024-activations-layer23", split="train")
+```
+
+Each example contains:
+```python
+{
+    "activations": np.ndarray,  # shape (1024, 3072), float32 (cast from bfloat16)
+    "mask": np.ndarray,         # shape (1024,), int32
+    "tokens": np.ndarray,       # shape (1024,), int32
+}
+```
+
+### Streaming (recommended for large datasets)
+
+```python
+ds = load_dataset(
+    "veriga/openwebtext-gemma3-tokenized-1024-activations-layer23",
+    split="train",
+    streaming=True
+)
+
+for example in ds:
+    activations = example["activations"]  # (1024, 3072)
+    mask = example["mask"]                # (1024,)
+    tokens = example["tokens"]            # (1024,)
+    # Use for Titans training...
+```
+
+### Manual loading (without HF datasets)
+
+```python
+import numpy as np
+
+shard_idx = 0
+activations = np.load(f"shard_{shard_idx:06d}.npy")        # (64, 1024, 3072), bfloat16
+mask = np.load(f"shard_{shard_idx:06d}_masks.npy")          # (64, 1024), int32
+tokens = np.load(f"shard_{shard_idx:06d}_tokens.npy")        # (64, 1024), int32
+```
+
+## Use Case: Titans Memory Layer
+
+This dataset is intended for training a [Titans](https://arxiv.org/abs/2501.00663) long-term memory module to be inserted after layer 23 of Gemma 3. The precomputed activations allow training the memory layer independently without running the full model forward pass.
+
+## Notes
+
+- Activations are stored in `bfloat16`; the HF Datasets loader casts them to `float32` for compatibility
+- Padding positions in activations are zeroed out via attention mask multiplication
+- The `metadata.json` file contains `{"next_shard": 1120}` used for resume during precomputation

openwebtext-gemma3-tokenized-1024-activations-layer23.py

@@ -0,0 +1,48 @@
+import numpy as np
+import glob
+import datasets
+
+
+_DESCRIPTION = """\
+Precomputed hidden state activations from layer 23 of Gemma-3-1B-IT
+for the OpenWebText dataset, tokenized with sequence length 1024.
+Designed for training a Titans memory layer.
+"""
+
+_SHARD_BASE = "shard_[0-9]*.npy"
+
+
+class OpenwebtextGemma3Tokenized1024ActivationsLayer23(datasets.GeneratorBasedBuilder):
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features({
+                "activations": datasets.Array3D(shape=(1024, 3072), dtype="float32"),
+                "mask": datasets.Array2D(shape=(1024,), dtype="int32"),
+                "tokens": datasets.Array2D(shape=(1024,), dtype="int32"),
+            }),
+        )
+
+    def _split_generators(self, dl_manager):
+        all_npy = sorted(glob.glob(_SHARD_BASE))
+        shards = [f for f in all_npy
+                  if not f.endswith("_masks.npy") and not f.endswith("_tokens.npy")]
+        return [datasets.SplitGenerator(
+            name=datasets.Split.TRAIN,
+            gen_kwargs={"shard_files": shards},
+        )]
+
+    def _generate_examples(self, shard_files):
+        for shard_path in shard_files:
+            shard_id = shard_path.split("shard_")[1].split(".npy")[0]
+            act = np.load(shard_path).astype(np.float32)
+            mask = np.load(shard_path.replace(".npy", "_masks.npy"))
+            tokens = np.load(shard_path.replace(".npy", "_tokens.npy"))
+            for i in range(act.shape[0]):
+                yield f"{shard_id}_{i}", {
+                    "activations": act[i],
+                    "mask": mask[i],
+                    "tokens": tokens[i],
+                }