* Model: More extensive checks for paged attention
Previously, TabbyAPI only checked for whether the user's hardware supports flash attention before deciding whether to enabled paged mode.
This adds checks for whether no_flash_attention is set, whether flash-attn is installed, and whether the installed version supports paged attention.
* Tree: Format
* Tree: Lint
* Model: Check GPU architecture first
Check GPU arch prior to checking whether flash attention 2 is installed
Cache size must be a multiple of 256 to work properly in ExllamaV2.
Take the config value and set the cache size to one multiple above
the remainder of the cache size divided by 256.
This is because cache size can never be lower than max_seq_len.
If max_seq_len isn't a multiple of 256, this method will never
yield a number that's lower than max_seq_len since it's no longer
a source of truth.
Signed-off-by: kingbri <bdashore3@proton.me>
Depending on the day of the week, Starlette can work with a CancelledError
or using await request.is_disconnected(). Run the same behavior for both
cases and allow cancellation.
Streaming requests now set an event to cancel the batched job and break
out of the generation loop.
Signed-off-by: kingbri <bdashore3@proton.me>
If a user is using GPU split, check compute capability on only those
GPUs. Autosplit assumes that all GPUs will be used.
Signed-off-by: kingbri <bdashore3@proton.me>
List comprehensions are the more "pythonic" way to approach mapping
values to a list. They're also more flexible across different collection
types rather than the inbuilt map method. It's best to keep one convention
rather than splitting down two.
Signed-off-by: kingbri <bdashore3@proton.me>
FA2 v2.5.7 and up is not supported below ampere and on AMD GPUs.
Clarify the error message and explain what happens as a result.
Signed-off-by: kingbri <bdashore3@proton.me>
Add a sequential lock and wait until jobs are completed before executing
any loading requests that directly alter the model. However, we also
need to block any new requests that come in until the load is finished,
so add a condition that triggers once the lock is free.
Signed-off-by: kingbri <bdashore3@proton.me>
The dynamic generator requires Flash attention 2.5.7 or higher to
be installed. This is only supported on Nvidia's 30 series and higher.
If a card is AMD or lower than the 30 series, switch to compatability
mode which functions the same way as the older generator, except
without parallel batching and any features that depend on it, such as
CFG.
Signed-off-by: kingbri <bdashore3@proton.me>
Dynamic generator needed multiple prompts to be tokenized and sent
for them to be sampled in serial, but generated in parallel.
Signed-off-by: kingbri <bdashore3@proton.me>
The new async dynamic job allows for native async support without the
need of threading. Also add logprobs and metrics back to responses.
Signed-off-by: kingbri <bdashore3@proton.me>
Dynamic gen takes in filters differently. Adjust to set the filter list
per class rather than in the generation function.
Signed-off-by: kingbri <bdashore3@proton.me>
Adds basic support for ExllamaV2's dynamic generator. Can generate
a streaming and non-streaming completion.
Signed-off-by: kingbri <bdashore3@proton.me>
skip_special_tokens is in stable exl2. Also default the parameters
if they are not present in the function signature.
Signed-off-by: kingbri <bdashore3@proton.me>
From exllamav2: List of strings that the generator will refuse to output. As soon as a partial match happens, a checkpoint is saved that the generator can rewind to if need be. Subsequent tokens are then held until the full string is resolved (match or no match) and either emitted or discarded, accordingly.
Bans the EOS token until the generation reaches a minimum length. This will not prevent the model from otherwise ending the generation early by outputting other stop conditions.
Appends the banned tokens to the generation. This is equivalent of
setting logit bias to -100 on a specific set of tokens.
Signed-off-by: kingbri <bdashore3@proton.me>
response_prefix is used to add a prefix before generating the next
message. This is used in many cases such as continuining a prompt
(see #96).
Also if a template has BOS token specified, add_bos_token will
append two BOS tokens. Add a check which strips a starting BOS token
from the prompt if it exists.
Signed-off-by: kingbri <bdashore3@proton.me>
Torch errors if float values are passed (because bytes are not float
types). Therefore, overestimate and cast to an int type.
Resolves#97
Signed-off-by: kingbri <bdashore3@proton.me>
Having many utility functions for initialization doesn't make much sense.
Instead, handle anything regarding template creation inside the
class which reduces the amount of function imports.
Signed-off-by: kingbri <bdashore3@proton.me>
GenerationConfig is meant to override various parts of the model
on generation within the transformers lib. Rather than implementing
the entire GenerationConfig framework (since it's pretty redundant),
add in multi eos_token support like VLLM.
The GenerationConfig is used only for generation, but can be used
for other uses if needed.
If there's more necessary parameters in the future, add those in
as well.
Signed-off-by: kingbri <bdashore3@proton.me>
LM format enforcer has some latency on token ingestion, so use an
optimized fork instead. Also add this in as a base dependency since
the size is small.
Signed-off-by: kingbri <bdashore3@proton.me>
Wrong class attribute name used for max_attention_size and fixes
declaration of the draft model's chunk_size.
Also expose the parameter to the end user in both config and model
load.
Signed-off-by: kingbri <bdashore3@proton.me>
HuggingFace updated transformers to provide templates in a list for
tokenizers. Update to support this new format. Providing the name
of a template for the "prompt_template" value in config.yml will also
look inside the template list.
In addition, log if there's a template exception, but continue model
loading since it shouldn't shut down the application.
Signed-off-by: kingbri <bdashore3@proton.me>
