Commit graph

1062 commits

Author SHA1 Message Date
Georgi Gerganov 5439a0ab57
ci : pip install gguf in editable mode (#2782)
ggml-ci
2023-08-25 13:03:25 +03:00
M. Yusuf Sarıgöz 8194cd8772
gguf : export objects to user code (#2780)
* gguf export more objects to user code

* gguf export all objects to user code for now

* gguf : bump version
2023-08-25 12:43:41 +03:00
Henri Vasserman 6bbc598a63
ROCm Port (#1087)
* use hipblas based on cublas
* Update Makefile for the Cuda kernels
* Expand arch list and make it overrideable
* Fix multi GPU on multiple amd architectures with rocblas_initialize() (#5)
* add hipBLAS to README
* new build arg LLAMA_CUDA_MMQ_Y
* fix half2 decomposition
* Add intrinsics polyfills for AMD
* AMD assembly optimized __dp4a
* Allow overriding CC_TURING
* use "ROCm" instead of "CUDA"
* ignore all build dirs
* Add Dockerfiles
* fix llama-bench
* fix -nommq help for non CUDA/HIP

---------

Co-authored-by: YellowRoseCx <80486540+YellowRoseCx@users.noreply.github.com>
Co-authored-by: ardfork <134447697+ardfork@users.noreply.github.com>
Co-authored-by: funnbot <22226942+funnbot@users.noreply.github.com>
Co-authored-by: Engininja2 <139037756+Engininja2@users.noreply.github.com>
Co-authored-by: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
Co-authored-by: jammm <2500920+jammm@users.noreply.github.com>
Co-authored-by: jdecourval <7315817+jdecourval@users.noreply.github.com>
2023-08-25 12:09:42 +03:00
Georgi Gerganov 3f460a2b72
cuda : add RoPE kernel for mode == 2 (NeoX) (#2760)
* cuda : add RoPE kernel for mode == 2 (NeoX)

* falcon : do not offload the embeddings layer
2023-08-25 11:55:59 +03:00
M. Yusuf Sarıgöz 87e3733f24
gguf : make gguf pip-installable
* gitignore : add dist and rm pyproject.toml

* gguf: prepare as Pip package

* gguf: prepare as Pip package

* gguf : fix line endings

* requirements : add gguf

* gguf : update readme with build notes

* gguf : update readme with build notes

* gguf : add notes for tests
2023-08-25 09:26:05 +03:00
Shouzheng Liu b91ad7f461
ggml-alloc : enlarge size of parse_seq (#2776)
Since we also store barriers in this array, we need to double its size.
2023-08-25 08:58:00 +03:00
Marcus Dunn 2e5f70a25f
Added enum to llama_token_get_type return type (#2774) 2023-08-24 23:49:30 +02:00
slaren d0f77b1353
convert.py : try to determine n_ctx automatically for CodeLlama (#2770) 2023-08-24 21:10:39 +02:00
slaren 0d3094f0c7
gguf : add rope_freq_base parameter for CodeLlama (#2769) 2023-08-24 21:04:05 +03:00
Georgi Gerganov 01f2224682
falcon : write file type 2023-08-24 19:58:30 +03:00
Shouzheng Liu 38b16dfca6
metal : bug-fix when enable ggml-alloc (#2757)
* metal: better memory alloc w/ concurrency dispatch

The ggml-alloc should only free tensors at memory barriers.

* ggml-alloc: avoid return silently

In certain cases, the allocate_node() function may silently return
without performing any memory allocation.
2023-08-24 19:27:25 +03:00
Georgi Gerganov 8f8c28e89c
convert : auto-determine model name based on dir + scripts update 2023-08-24 19:26:47 +03:00
Kerfuffle 7694adda8d
Fix for main example getting stuck when -n -2 and --interactive (#2767)
* Fix for main example getting stuck when -n -2 and --interactive

* Add a comment so future generations may suffer less.
2023-08-24 10:11:13 -06:00
slaren fea95c682d
fix convert.py for codellama, add llama 34B to the list of recognized models (#2768) 2023-08-24 17:44:11 +02:00
DannyDaemonic ef955fbd23
Tag release with build number (#2732)
* Modified build.yml to use build number for release

* Add the short hash back into the tag

* Prefix the build number with b
2023-08-24 15:58:02 +02:00
Georgi Gerganov d67777c202
metal : add Q8_0 support (#2763)
* metal : add dequantize_q8_0 kernel

* metal : add mul_mat_q8_0_f32 kernel

* metal : add Q8_0 mul_mm kernel
2023-08-24 16:19:57 +03:00
Georgi Gerganov c3e53b421a
llama : escape all U+2581 in a string (#2750) 2023-08-24 12:26:01 +03:00
Evan Jones 6e91a1b070
llama : fix grammar sometimes generating null char (#2756) 2023-08-24 07:07:13 +03:00
Georgi Gerganov 44d5462b5c
readme : fix link 2023-08-23 23:44:19 +03:00
Georgi Gerganov c7868b0753
minor : fix trailing whitespace 2023-08-23 23:43:00 +03:00
Georgi Gerganov 79da24b58c
readme : update hot topics 2023-08-23 23:41:16 +03:00
Georgi Gerganov cf658adc83
llm : add Falcon support (#2717)
* llama : refactor GGUF constants into static maps

* llama : check if model architecture is known

* llama : refactor llama_model_load_internal()

* gguf : add KV constant maps

* llm : read arch-specific KVs

* convert : add dummy scores + types

* falcon : load tensor data (CPU only)

* llama : fix loading progress bar

* llama : add arch member to llama_model

* falcon : CPU inference working

* falcon : support non-40B models

* falcon : minor

* llama : minor updates

ggml-ci

* convert-falcon-hf-to-gguf.py : fix special token mapping

* llama.cpp : llama default UNK token = id 0

* llama.cpp : fix bpe tokenizer

* llama.cpp : fix the fix of bpe tokenizer

* ggml : pass eps to ggml_norm

* metal : implement RoPE (mode = 2) + avoid ggml_repeat

* ggml : ggml_repeat always creates new tensor

* falcon : copy-paste self-attention from LLaMA

* metal : print extra compute pipeline info

* falcon : minor changes (still chasing the Metal problem)

* llama.cpp : fix linefeed token

* metal : fix GELU kernel numerical stability by using precise::tanh

* metal : temporary workaround for the concurrency optimization bug

* falcon : add CUDA offloading (#2739)

* llama : better model naming and size reporting

* llama : prep new tokenizer support

* llama : advanced BPE tokenizer based on ggllm.cpp imlpementation

* llama : remove oboslete comment

ggml-ci

* common : remove obsolete BPE API + disable test-tokenizer-1

* llama : revert BPE special-case in llama_byte_to_token()

* cuda : add TODOs for RoPE NeoX implementation

* llama : default special tokens based on vocab type

* perplexity : add log for start of tokenization

---------

Co-authored-by: klosax <131523366+klosax@users.noreply.github.com>
Co-authored-by: slaren <slarengh@gmail.com>
2023-08-23 23:08:04 +03:00
Georgi Gerganov a192860cfe
minor : fix trailing whitespace 2023-08-23 22:37:39 +03:00
Olivier Chafik 95385241a9
examples : restore the functionality to import llama2.c models (#2685)
* Fix import of llama2.c models that don't share weights between embedding layers

* llama2c: reinstate ggmlv3 conversion output + update readme w/ gguf conv

* llama2.c: comment out legacy "load from ggml model" logic

* llama2.c: convert special-cased "<0xXX>" single byte tokens from tokenizer.bin
2023-08-23 22:33:05 +03:00
slaren 335acd2ffd
fix convert-lora-to-ggml.py (#2738) 2023-08-23 16:46:54 +02:00
klosax 5290c38e6e
main : insert bos if no tokens (#2727)
* main.cpp : insert bos if no tokens

* Update examples/main/main.cpp

* Update examples/main/main.cpp

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-08-23 16:46:03 +02:00
akawrykow cc34dbda96
gitignore : fix for windows (#2729) 2023-08-23 17:31:34 +03:00
Cebtenzzre 7c2227a197
chmod : make scripts executable (#2675) 2023-08-23 17:29:09 +03:00
JohnnyB f19dca04ea
devops : RPM Specs (#2723)
* Create llama-cpp.srpm

* Rename llama-cpp.srpm to llama-cpp.srpm.spec

Correcting extension.

* Tested spec success.

* Update llama-cpp.srpm.spec

* Create lamma-cpp-cublas.srpm.spec

* Create lamma-cpp-clblast.srpm.spec

* Update lamma-cpp-cublas.srpm.spec

Added BuildRequires

* Moved to devops dir
2023-08-23 17:28:22 +03:00
Kawrakow 8207214b6a
Fix values shown in the quantize tool help (#2735)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-23 12:57:12 +03:00
Kawrakow 62959e740e
Strided perplexity (#2714)
* Implementing strided computation of perplexity

* Alternative way to output PPL results

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-23 12:56:42 +03:00
IgnacioFDM 7f7ddd5002
Fix ggml to gguf conversion on Windows (#2733)
This fixes `RuntimeWarning: overflow encountered in long_scalars`

Credit: anon (not mine)
2023-08-23 03:31:09 -06:00
Xiao-Yong Jin b8ad1b66b2
server : allow json array in prompt or content for direct token input (#2306)
* server: allow json array in prompt or content

We accept an array of strings and numbers representing tokens,
in addition to the current string valued prompt or content.

This allows direct token input, so that any special tokens
can be processed and used at the frontend during the construction
of the json data, before sending to the server. And the server
does not need to know or parse special tokens from textual input.

With this, we can use EOS and BOS used in llama-2-chat models.

* server: use tokenizePrompt(json) and default "" if empty prompt

* server: fix prompt check

* server: tokenize endpoint no longer adds BOS
2023-08-23 15:12:12 +08:00
Evan Jones f5fe98d11b
docs : add grammar docs (#2701)
* docs : add grammar docs

* tweaks to grammar guide

* rework GBNF example to be a commented grammar
2023-08-22 21:01:57 -04:00
Kerfuffle 777f42ba18
Improve handling of special tokens in GGML to GGUF converter (#2725)
* Improve UNK, BOS, EOS token handling when converting without metadata.

* Allow importing as a module.

* Remove some obsolete code and minor cleanups.

* Set default UNK token mapping from -1 to 0 in llama.cpp

* Try to handle overflow due to buggy Windows Python with a better error message
2023-08-22 17:39:39 -06:00
goerch 46ef5b5fcf
llama : fix whitespace escaping in tokenizer (#2724) 2023-08-23 00:10:42 +03:00
Johannes Gäßler c63bb1d16a
CUDA: use mul_mat_q kernels by default (#2683) 2023-08-22 22:47:05 +02:00
Alex Petenchea 3b6cfe7c92
convert.py : clarifying error message (#2718) 2023-08-22 21:58:16 +03:00
Jiahao Li 800c9635b4
Fix CUDA softmax by subtracting max value before exp (#2665) 2023-08-22 20:27:06 +02:00
Georgi Gerganov deb7dfca4b
gguf : add ftype meta info to the model (#2710)
* llama : add ftype meta info to the model

ggml-ci

* convert.py : add ftype when converting (does not work)

* convert.py : fix Enum to IntEnum

ggml-ci
2023-08-22 20:05:59 +03:00
Kawrakow bac66994cf
Quantization imrovements for k_quants (#2707)
* Improve LLaMA-2 2-, 3- and 4-bit quantization

* Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2
* Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2
* Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of
  attention.wv and feed_forward.w2

This leads to a slight model sized increase as follows:
Q2_K  : 2.684G vs 2.670G
Q3_K_S: 2.775G vs 2.745G
Q3_K_M: 3.071G vs 3.057G
Q4_K_S: 3.592G vs 3.563G

LLaMA-2 PPL for context 512 changes as follows:
Q2_K  : 6.6691 vs 6.8201
Q3_K_S: 6.2129 vs 6.2584
Q3_K_M: 6.0387 vs 6.1371
Q4_K_S: 5.9138 vs 6.0041

There are improvements for LLaMA-1 as well, but they are
way smaller than the above.

* Minor 4-bit quantization improvement

For the same model size as previus commit, we get
PPL = 5.9069 vs 5.9138.

* Some more fine tuning

* Adding make_qkx2_quants

With it, we get PPL = 5.8828 for L2-7B Q4_K_S.

* Another minor improvement

* Q2_K improvement

Smaller model, lower perplexity.
 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201
12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178

It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk,
which are Q2_K

* Iterating

* Revert Q5_K back to make_qkx1_quants

* Better Q6_K

* make_qkx2_quants is better for Q5_K after all

* Fix after rebasing on master

* Fix for changed tensor names

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2023-08-22 19:14:09 +03:00
slaren 519c981f8b
embedding : evaluate prompt in batches (#2713) 2023-08-22 16:03:12 +02:00
slaren 1123f7fbdf
ggml-cuda : use graph allocator (#2684)
use a different function for no_alloc to avoid breaking backwards compat, fixes lora

remove 512 n_batch limit

fixed 2048 batch size

cleanup

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
2023-08-22 15:25:19 +02:00
Georgi Gerganov ef3f333d37
ggml : sync latest (SAM + SD operators, CUDA alibi) (#2709)
* ggml : sync latest (SAM + SD operators, CUDA alibi)

ggml-ci

* ggml : fix tabs
2023-08-22 14:22:08 +03:00
slaren 8e4364f2af
llama-bench : minor fixes (#2695) 2023-08-22 10:56:03 +03:00
Kylin 1e3bc523d8
ggml : support CUDA's half type for aarch64(#1455) (#2670)
* ggml: support CUDA's half type for aarch64(#1455)
support CUDA's half type for aarch64 in ggml_fp16_t definition

* ggml: use __CUDACC__ to recognise nvcc compiler
2023-08-22 10:14:23 +03:00
Shouzheng Liu 14b1d7e6f7
metal : add missing barriers for mul-mat (#2699) 2023-08-22 09:18:40 +03:00
Jhen-Jie Hong 226255b44e
server : fallback to default if client param is null (#2688)
* server : fallback to default if client param is null

* server : do not overwrite 404 if status is 500 from exception_handler
2023-08-22 08:32:00 +08:00
Kerfuffle 930523c8e1
Fix convert-llama-ggmlv3-to-gguf.py vocab conversion (#2698)
When converting without metadata, the hex value for bytes entries weren't 0 padded to 2 digits.
2023-08-21 18:01:34 -06:00
Georgi Gerganov c8dba409e6
py : remove obsolete script 2023-08-21 23:40:22 +03:00