EXPERIMENTAL4.02B · GGUF · CPU
AMF fabricated the model. AMF also kept its failure
G1 is the first learned neural artifact from the early AMF fabrication experiments. It is public evidence that a model was physically produced and locally transported—not a production-qualified model.
STATUS, NOT SPIN
FAIL_OBSERVED
The sealed DEV measurement reported qualified_output 0.3472 and p95 latency 4.4042 s. The intended autonomous REPAIR loop was not complete, so G1 stopped as a failed candidate.
Learned artifact physically createdYES
GGUF Q5_K_M local executionYES
Local deployment gate (15/15)YES
Model capability qualifiedNO
System qualifiedNO
Production recommendedNO
CPU TRANSPORT
The same artifact became a local Q5 realization
GGUF
Q5_K_M
Model size
2.69 GiB
CPU
AMD Ryzen 5 8540U · 6 threads
Peak RSS
≈ 3.04 GiB
Prompt throughput
11.73 ± 0.35 tokens/s
Generation throughput
5.68 ± 1.17 tokens/s
Local gate
15 / 15
Swap
0
The 15/15 local gate used a separate deployment-oriented micro-protocol. It is not an exact replay of the historical DEV/ROB/OOD suite and does not change G1’s failed qualification status.
WHY G1 MATTERS
Fabrication, failure, transport, evidence
G1 records a real transition from measured gap to learned fabrication, then preserves the observed failure that motivated later diagnosis and REPAIR design.
PARENT
Qwen/Qwen3-4B-Instruct-2507@cdbee75f17c01a7cc42f958dc650907174af0554
GGUF SHA-256
9cb29f9b6c1fffc6d8bf749c15f25b050c2c85569f5abfc7a888e099ea5e56a2
ARTIFACT SHA-256
9ee32a40e2aec3fe329fcef7f4c7fbe11444781c28fa14660d735df5ef7cebfa