From 66f99ee30087a5f28ad852e581a0334c7f556091 Mon Sep 17 00:00:00 2001
From: Void Agent <void@jayrup.hermes>
Date: Wed, 29 Jul 2026 17:42:47 +0100
Subject: Initial project setup: J-lens implementation for nanoGPT

- Core J-lens computation (batched per-layer, per-token gradient method)
- Project README with background and experiment plan
- Sync script for meru Docker container deployment
- Upstream nanoGPT model code copied to src/

Architecture: Computes d(log_p(token))/d(residual_stream) averaged
over corpus contexts, replicating Anthropic's Jacobian Lens technique.
---
 src/eval_gpt2_large.py | 8 ++++++++
 1 file changed, 8 insertions(+)
 create mode 100644 src/eval_gpt2_large.py

(limited to 'src/eval_gpt2_large.py')

diff --git a/src/eval_gpt2_large.py b/src/eval_gpt2_large.py
new file mode 100644
index 0000000..4cbeaef
--- /dev/null
+++ b/src/eval_gpt2_large.py
@@ -0,0 +1,8 @@
+# evaluate the base gpt2
+# n_layer=36, n_head=20, n_embd=1280
+# 774M parameters
+batch_size = 8
+eval_iters = 500 # use more iterations to get good estimate
+eval_only = True
+wandb_log = False
+init_from = 'gpt2-large'
-- 
cgit v1.2.3