Timmy_Foundation/timmy-config
16
0
Fork 0
Code Issues 57 Pull Requests 1 Actions 1 Packages Projects Releases Wiki Activity

v1.0 rejected — NaN from wrong tokenizer, Morrowind MCP pipeline working

Browse Source
This commit is contained in:
Alexander Whitestone
2026-03-26 12:32:08 -04:00
parent 16675abd79
commit f9155b28e3
1 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
autolora/manifest.yaml
View File

@@ -55,7 +55,8 @@ adapters:
timmy-v1.0: timmy-v1.0:
base: hermes4-14b-4bit base: hermes4-14b-4bit
date: 2026-03-26 date: 2026-03-26
status: training status: rejected
data: 1125 train / 126 valid (same curated set, reused) data: 1125 train / 126 valid (same curated set, reused from 8B — NOT re-tokenized)
training: { lr: 1e-6, rank: 16, iters: 800 } training: { lr: 1e-6, rank: 16, iters: 800 }
notes: "First 14B adapter. Conservative lr for new arch." eval: "Val NaN iter 100, train NaN iter 160. Dead."
notes: "Data was pre-truncated for Llama3 tokenizer, not Qwen3. Must re-run clean_data.py with 14B tokenizer before v1.1."