04c91d29ff
use ggml_format_name
2024-06-11 19:14:04 +02:00
54f77e2467
add to makefile all targets
2024-06-11 19:03:13 +02:00
3223133cf5
default n_pca_batch to 20
2024-06-11 15:05:06 +02:00
d41c719980
bring back n_completions
2024-06-11 14:31:45 +02:00
446da906d9
fix n_completions
2024-06-11 08:22:38 -04:00
1a088fb0a5
working version
2024-06-11 12:37:05 +02:00
9e39571fc2
add n_batch for pca
2024-06-11 11:45:16 +02:00
6a5adf3d7c
fix shape of v_diff_original
2024-06-11 01:33:16 +02:00
c241b500a1
clean up PCA ggml implementation
2024-06-11 01:13:10 +02:00
a710df749c
(wip) refactor
2024-06-07 15:37:58 +02:00
a42e783d75
update comments
2024-06-03 21:33:46 -04:00
3815a0c306
pre-tokenize so we can allocate correct memory to ctx_diffs_wrapped
2024-06-03 21:26:13 -04:00
23fd1b587c
update debug statements
2024-06-03 21:14:43 -04:00
07dba13ab6
temporary commit while I move dev environments
...
it finally outputs a functioning control vector - "functioning" in the sense that it can be loaded and it clearly has the right idea, but makes the model incoherent
2024-06-03 17:40:19 -04:00
15d5c257a0
fix cb_eval
2024-06-02 10:58:11 +02:00
a23c72e4c0
fix ggml errors and make new ones
...
at least it compiles and runs
2024-06-01 22:19:33 -04:00
b67ea65983
tentatively translate the rest
2024-06-01 20:47:28 -04:00
0e1f9734de
translated everything but PCA (I think)
2024-06-01 19:50:46 -04:00
df623fffe8
interim fix memory leak
2024-06-01 18:36:54 -04:00
3090c485b6
remove unnecessary multithreading
2024-06-01 18:32:14 -04:00
544268888b
in-series multithreading for prompt embedding?
...
added commented-out code to attempt to start implementing mutlithreading for embedding in main
2024-06-01 17:25:21 -04:00
86842b20e5
fix compiler warnings
2024-05-31 22:25:46 -04:00
db3ba108e7
code aestheticization
2024-05-31 21:38:02 -04:00
62560367aa
add command-line args for num threads, num completions file lines, always reload model
...
refactored a few things and did what the commit message says on the tin
2024-05-31 21:27:14 -04:00
4d7d71bc43
fix square_diff matmul index range and CRLF->LF line endings
...
fixed a logic error where square_diff would not multiply all rows
fixed a formatting error where the provided completions.txt had CRLF line endings
2024-05-31 21:08:25 -04:00
4d88cd1af1
fix zero output & param parsing, functional templating
...
fixed a bug where the output file had no tensor data/was all zero
fixed a bug where single hyphen flags were not being correctly parsed
implements creation of templated prompts from input (still need to adapt based on model)
2024-05-31 12:40:35 -04:00
fa85ba6ae3
preliminary template/multiprompt support
...
model is running out of context and that ought to be fixed (segfaulting) but other than that it looks goodish
2024-05-30 23:39:59 -04:00
31f153fe9c
fix matrix transpose multiplication
...
you have got to be kidding me
2024-05-30 21:36:17 -04:00
d446c6d887
add debugs
2024-05-31 00:41:12 +02:00
287da25f48
fix mem error
2024-05-31 00:06:45 +02:00
447023fc43
add multi prompts, multi-thread for PCA
2024-05-30 23:58:32 +02:00
dc46264ff0
example template completions
...
Implements an example template set built from the positive/negative prompts like the control vector Python implementation.
2024-05-30 13:12:54 -04:00
f58f6af133
param parsing, refactor, comments
...
Added basic command-line parameters for outfile and one each positive/negative prompt.
Refactored some messy code in PCA computation and GGUF exporting.
Left a bunch of comments regarding further work needed.
2024-05-30 11:31:45 -04:00
73747fe8eb
proof-of-concept stdlib implementation
...
Implements PCA and file writing using mostly standard libraries. The output is recognized as a functional control vector, but outputs gibberish.
2024-05-30 00:31:29 -04:00
b30bea3257
add comments
2024-05-24 22:50:03 +02:00
c31c118d86
calc diff
2024-05-24 11:46:47 +02:00
0a46d73056
add control-vector-generator
2024-05-24 11:11:55 +02:00
ngxson
Christian Zhou-Zheng