Upload lact 350M r64 (10K steps, FineWeb-Edu)

README.md

@@ -0,0 +1,65 @@
+---
+license: apache-2.0
+tags:
+- low-rank-fast-weights
+- linear-attention
+- lact
+- r64
+- 350M
+datasets:
+- HuggingFaceFW/fineweb-edu
+---
+
+# LaCT (Large-Chunk Test-Time Training, SwiGLU MLP fast weights) 350M (rank 64) — Low-rank Fast-Weight Ablation
+
+Pretrained 350M-parameter LaCT (Large-Chunk Test-Time Training, SwiGLU MLP fast weights) with low-rank parameterization
+(`r64`) on FineWeb-Edu. Part of a multi-cell ablation across
+4 archs × {`r32`, `r64`, `r256`, `rfull`} (plus GDN extras `r512`)
+studying whether constraining the q/k/v fast-weight projections (or LaCT's
+SwiGLU MLP) to low rank can match or exceed full-rank performance at the
+350M scale.
+
+## Training
+
+| | |
+|---|---|
+| Architecture | LaCT (Large-Chunk Test-Time Training, SwiGLU MLP fast weights) |
+| Rank | `r64` |
+| Params | ~350M (hidden=1024, layers=24, heads=16) |
+| Dataset | `HuggingFaceFW/fineweb-edu` (streaming) |
+| Steps | 10000 |
+| Effective batch | 256 |
+| Sequence length | 8000 |
+| Optimizer | AdamW (lr=3e-4, eps=1e-15) |
+| LR schedule | Cosine, 512-step warmup, decay to 10% |
+| Precision | bf16 |
+| Activation checkpointing | selective (option 1) |
+| Tokens | ~20.5 B |
+
+## Eval results
+
+- **FineWeb-Edu val PPL**: `13.42`
+- **LAMBADA acc**: 0.297
+- **HellaSwag acc_norm**: 0.367
+- **ARC-Easy acc_norm**: 0.470
+- **ARC-Challenge acc_norm**: 0.254
+- **PIQA acc_norm**: 0.648
+- **WinoGrande acc**: 0.520
+
+
+## Notes on the 350M sweep
+
+- **Downstream eval discrimination comes online at 350M.** At 100M, HellaSwag /
+  LAMBADA were near-chance for most cells; at 350M they discriminate clearly
+  between archs/ranks.
+- **PPL doesn't linearly predict downstream.** At matched ~374M, GLA `rfull`
+  has worse FineWeb-Edu PPL than DeltaNet `rfull` (14.42 vs 12.55) but wins on
+  *every* lm-harness task (LAMBADA, HellaSwag, PIQA, ARC-E).
+- **GatedDeltaNet dominates at the cost of size.** GDN `rfull` is ~526M
+  (head_dim=256 inflates q/k/v) and wins every metric; GDN `r256` (~432M) is
+  the matched-param comparison and still leads.
+- **LaCT is rank-robust at 350M.** PPL/LAMBADA stay flat across r64 / r256 /
+  rfull — the cleanest evidence for the "low rank as regularization"
+  hypothesis.
+
+Run name: `lact_350M_r64_bs256_lr3e-4_steps10000`

config.json

@@ -0,0 +1,50 @@
+{
+  "architectures": [
+    "LaCTForCausalLM"
+  ],
+  "attention_bias": false,
+  "attn_qk_norm": false,
+  "bos_token_id": 1,
+  "dtype": "float32",
+  "elementwise_affine": true,
+  "eos_token_id": 2,
+  "fp32_states": false,
+  "fuse_cross_entropy": true,
+  "fuse_norm": true,
+  "fuse_swiglu": true,
+  "fw_init_gain": 0.5,
+  "hidden_act": "swish",
+  "hidden_ratio": 4,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "inter_multi": 1,
+  "intermediate_size": null,
+  "lact_chunk_size": 2048,
+  "last_layer_fuse_norm": true,
+  "learnable_ttt_scale": true,
+  "lr_dim": 1,
+  "lr_parameterization": "mamba",
+  "max_position_embeddings": 8192,
+  "model_type": "lact_swiglu",
+  "no_v_silu": false,
+  "norm_eps": 1e-06,
+  "num_attn_heads": 16,
+  "num_heads": 8,
+  "num_hidden_layers": 24,
+  "num_lact_heads": 2,
+  "qkv_bias": false,
+  "qkv_silu": true,
+  "rope_theta": 1000000,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.6",
+  "ttt_loss_type": "dot_product",
+  "ttt_nope": false,
+  "ttt_prenorm": true,
+  "use_cache": true,
+  "use_fused_kernel": false,
+  "use_momentum": true,
+  "use_muon": false,
+  "vocab_size": 32000,
+  "w0_w2_low_rank": 64,
+  "window_size": 2048
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "4.57.6"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99bd8b5a98428dcd1b0280ba2ba48985e042c8288cca24098a69d01245ece993
+size 1572448336

special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

tokenizer_config.json

@@ -0,0 +1,44 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}