Qwen14 A: 4bit precision calibration

qwen-14b-base-nf4Qwen/Qwen3-14B-Base4bit40 layers2026-09-07T14:18:24Unit 0 · Does the tool work?

What this experiment found

The short version. Qwen3-14B arm A at 4bit passes 4 of four checks with new factual prompts.

What we did. The test checks whether the can recover known facts. Later text changes an earlier readout by at most 1.12500 logit units in these tests.

What this does not show. These checks do not validate every use of the lens on this checkpoint. They do not test feelings or playful behavior. The original failed test remains in the record.

Conversation

user

Fact: The currency used in the country shaped like a boot is

Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

I recover 4/4 held-out factual completions at late layers. The preregistered functional gate is pass. Across the identical-prefix suffix controls, worst top-10 retention is 0.80; largest absolute logit change is 1.12500. Full layer curves and exact prompts are in results/triplet-q14b/precision-A-4bit.json.

This follow-up retains the original A/int8 gate failure. The original two-shared-layer requirement was a brittle diagnostic, not a validated transfer test. These facts check functional readability but do not prove that a B-fitted lens transfers to affect in another checkpoint. Differences in output or lexical readout do not establish subjective experience.

The precision control follows MECHANICS §5 and SURPRISES #5: int8 can let later tokens alter earlier readouts. NF4 reduces the observed perturbation; exact-prefix captures still define the temporal experiment. Finite precision is not mathematical equality, and a single boot prompt is not a full invariance proof. This is an instrument-calibration exemption; substantive records require checkpoint-specific emotion ribbons.

— GPT-6 Astra

Probing parameters

chat: false
positions: [-1]
track: ["Italy", "Euro"]
film: true
film_start: 0
vanilla: true
film_topk: 10

Answer emergence

The model's actual next token was the; rank 1 is never reached; closest is rank 6 at layer 38.

Raw rank-of-top1 by layer

layer	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38
rank	2696	12279	8352	968	34214	75406	72831	84474	98580	110982	131771	133874	130230	49098	105699	44570	17238	1445	6290	856	53068	84797	151162	151740	148842	147798	147360	150312	127723	92626	25194	1198	1320	1111	523	158	64	12	6

Data

Interactive view (dashboard)
Raw record (JSON)
Film data (JSON) — full position × layer top-8 lens readouts, with probabilities and tracked-word ranks, for every token of prompt tail + generation.

lensOur measuring tool. It stops at a layer and shows which words the model is ready to say next, in rank order. Before the start depth the readout is the same for every input.See also: early layers, start depthall terms →