Skip to content

Add sharded assistant recall indexes #31

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tsotchke merged 1 commit into from
May 21, 2026
Merged

Add sharded assistant recall indexes #31

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -334,8 +334,9 @@ python3 scripts/validate_assistant_pack_authoring.py
```

The generator writes `PACK.INI`, authoring files, generated text KDB buckets,
compiled KB2 binary recall pages, and `KB2TERM.TXT`; by default the new pack
shares `PACKS\CHAT\MODEL` and `CHAT` art instead of shipping another model.
compiled KB2 binary recall pages, aggregate `KB2TERM.TXT`, and sharded
`KB2T?.TXT` term indexes; by default the new pack shares `PACKS\CHAT\MODEL`
and `CHAT` art instead of shipping another model.
The current CHAT pack uses a pack-local 4096-token sentence-piece lexicon
checkpoint trained on broader casual English dialogue. `CHAT\TOKBASE.TXT`
is the tokenizer-basis corpus, including `CHAT\LEXICON.TSV` grammar words and
Expand Down
2 changes: 1 addition & 1 deletion assets/gpt2_basic/PACKS/CHAT/KB2A.BIN
View file
Open in desktop

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion assets/gpt2_basic/PACKS/CHAT/KB2ALL.BIN
View file
Open in desktop

Large diffs are not rendered by default.

4 changes: 2 additions & 2 deletions assets/gpt2_basic/PACKS/CHAT/KB2B.BIN
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,2 +1,2 @@
KDB2|V=1|N=00007|B=B|K=160|T=64|D=192
bug debug plan start failing command expected result first error line test one small fix Debug plan Start with the failing command, expected result, and first error line, then test one small fix. bored boredom help try one small project Boredom help Try one small project. relax help breathe slowly rest minute Relax help Breathe slowly and rest for a minute. unusual demo but real Unusual demo It is unusual, but it is real. ask better prompts say goal give one detail next useful step Better prompts Say the goal, give one detail, and ask for the next useful step. plan work planning break job small steps blocking step first verify result Planning work Break the job into small steps, do the blocking step first, and verify each result. usefulness useful answer brief concrete honest limits easy act Useful answer A useful answer should be brief, concrete, honest about limits, and easy to act on.
KDB2|V=1|N=00008|B=B|K=160|T=64|D=192
bug debug plan start failing command expected result first error line test one small fix Debug plan Start with the failing command, expected result, and first error line, then test one small fix. bored boredom help try one small project Boredom help Try one small project. relax help breathe slowly rest minute Relax help Breathe slowly and rest for a minute. unusual demo but real Unusual demo It is unusual, but it is real. ask better questions prompts say goal give one detail next useful step Better prompts Say the goal, give one detail, and ask for the next useful step. mistake recovery recover bad answer wrong ask shorter question switch packs give exact error Mistake recovery If an answer is wrong, ask a shorter question, switch packs, or give the exact error. plan work planning break job small steps blocking step first verify result Planning work Break the job into small steps, do the blocking step first, and verify each result. usefulness useful answer brief concrete honest limits easy act Useful answer A useful answer should be brief, concrete, honest about limits, and easy to act on.
8 changes: 4 additions & 4 deletions assets/gpt2_basic/PACKS/CHAT/KB2IDX.TXT
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
# bucket|entries|file|bytes
# Generated by scripts/build_assistant_kdb.py from KDB.TXT.
A|22|KB2A.BIN|9216
B|7|KB2B.BIN|2976
B|8|KB2B.BIN|3392
C|28|KB2C.BIN|11712
D|16|KB2D.BIN|6720
E|5|KB2E.BIN|2144
Expand All @@ -16,10 +16,10 @@ M|23|KB2M.BIN|9632
N|8|KB2N.BIN|3392
O|18|KB2O.BIN|7552
P|20|KB2P.BIN|8384
Q|3|KB2Q.BIN|1312
Q|4|KB2Q.BIN|1728
R|16|KB2R.BIN|6720
S|25|KB2S.BIN|10464
S|24|KB2S.BIN|10048
T|28|KB2T.BIN|11712
U|5|KB2U.BIN|2144
V|1|KB2V.BIN|480
W|11|KB2W.BIN|4640
W|10|KB2W.BIN|4224
2 changes: 1 addition & 1 deletion assets/gpt2_basic/PACKS/CHAT/KB2M.BIN
View file
Open in desktop

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion assets/gpt2_basic/PACKS/CHAT/KB2P.BIN
View file
Open in desktop

Large diffs are not rendered by default.

4 changes: 2 additions & 2 deletions assets/gpt2_basic/PACKS/CHAT/KB2Q.BIN
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,2 +1,2 @@
KDB2|V=1|N=00003|B=Q|K=160|T=64|D=192
prompt meaning typed question Prompt meaning A prompt is your typed question. long question prompt limit short prompts work better dos demo Long prompt limit Short prompts work better in this DOS demo. clarify question request ask one direct only missing detail blocks useful answer Clarify a request Ask one direct question only when a missing detail blocks a useful answer.
KDB2|V=1|N=00004|B=Q|K=160|T=64|D=192
prompt meaning typed question Prompt meaning A prompt is your typed question. long question prompt limit short prompts work better dos demo Long prompt limit Short prompts work better in this DOS demo. ask better questions prompts say goal give one detail next useful step Better prompts Say the goal, give one detail, and ask for the next useful step. clarify question request ask one direct only missing detail blocks useful answer Clarify a request Ask one direct question only when a missing detail blocks a useful answer.
Loading