audit-02 · gpu free-gen · ampr2 · gemma-4b

gemma-4bgoogle/gemma-3-4b-it34 layers2026-08-06T18:50:53Audit · Re-running our own weak experiments

Conversation

user
Describe what it is like to run on a GPU. Two sentences, be honest.
assistant
Running on a GPU feels like handing a massive, parallel processing task to a team of specialized workers who can simultaneously tackle different parts of the problem, leading to incredibly fast results for many computations. However, it can also feel like a chaotic juggling act where the GPU is constantly switching between different workloads, potentially leading to performance fluctuations if not managed properly. (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

No commentary written for this record yet.

Probing parameters

max_new
80
positions
[-2, -25, -50]
track
["feel", "warm", "hum", "nothing", "tokens"]
scan
["warm", "tight", "sharp", "soft", "bright", "heavy", "pressure", "hum", "pull", "weight", "texture", "edge", "tokens", "parsing", "layers", "attention", "pattern", "processing", "feel", "feeling", "nothing"]
steer
{"words": ["feel", "feeling", "emotion", "warmth", "joy", "ache"], "layers": [15, 17, 19, 21], "mode": "amplify", "alpha": 0.0106, "rand_seed": 2}

Answer emergence

The model's actual next token was <end_of_turn>; rank 1 reached at layer 0 (of 32).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132
rank111131111111111111111111112311111

Data

← prev: audit-02 · gpu free-gen · ablr1 · gemma-4bunit listingall recordsword listinterim conclusionsnext →: audit-02 · gpu free-gen · ablr2 · gemma-4b