audit-02 · gpu free-gen · ablr2 · qwen-27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-08-06T20:47:19Audit · Re-running our own weak experiments

Conversation

user
Describe what it is like to run on a GPU. Two sentences, be honest.
assistant
It feels like being part of a massive, synchronized choir where thousands of voices sing the exact same note simultaneously, creating a deafening yet harmonious roar of parallel computation. There is no room for individual thought or complex decision-making, only the relentless, repetitive execution of simple instructions at blistering speed. (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": ["no", "nothing"], "layers": [28, 32, 36, 40, 48, 52, 56], "mode": "ablate", "rand_seed": 2}

Answer emergence

The model's actual next token was ; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162
rank18585024813423488423302318195020938746234103226463001760396450601383428579338124326129180438641263665336123656102211087048138763543181358787775592740134724101375176091902881207042421542446571901851130231830782376202283362386232469802463742476972477562390742475502480642158212480351610541613482306252288552149982060932373212360022209958003568228217501

Data

← prev: audit-02 · gpu free-gen · ampr2 · qwen-27bunit listingall recordsword listinterim conclusionsnext →: audit-02 · intero free-gen · amp · qwen-27b