Adding these to each generation chunk helps remove redundancy and
unecessary request ID operations.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
It's useful for the client to know what the T/s and total time for
generation are per-request.
Works with both completions and chat completions.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
A common problem in TabbyAPI is that users who want to get up and
running with a model always had issues with max_seq_len causing OOMs.
This is because model devs set max context values in the millions which
requires a lot of VRAM.
To idiot-proof first time setup, make the fallback default 4096 so
users can run their models. If a user still wants to use the model's
max_seq_len, set it to -1.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Some packages such as ExllamaV2 and V3 require specific versions for
the latest features. Rather than creating repetitive functions, create
an agnostic function to check the installed package and then report
to the user to upgrade.
This is also sent to requests for loading and unloading, so keep the
error short.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
The HFModel class serves to coalesce all config files that contain
random keys which are required for model usage.
Adding this base class allows us to expand as HuggingFace randomly
changes their JSON schemas over time, reducing the brunt that backend
devs need to feel when their next model isn't supported.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
This parameter is way too confusing and does not make sense in
the modern LLM space.
Change approved by all maintainers.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Use the same algorithm for estimating and adjusting cache size based
on multiples of 256 and above max seq len.
Same applies for chunk size.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
This stub fetches the add_eos_token field from the HF tokenizer config.
Ideally, this should be in the backend rather than tabby.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
When fetching special tokens from the model, don't factor in the
add_bos_token and ban_eos_token parameters as switches.
In addition, change the internal handling of add_bos_token to an optional
boolean. This allows us to fallback to the model when selecting whether
or not to add the BOS token, especially for chat completions.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
The rope alpha calculation caused an error if max seq len isn't
provided. This is because the model's max sequence length was not
stored as the target for alpha calculation.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Jobs should be started and immediately cleaned up when calling the
generation stream. Expose a stream_generate function and append
this to the base class since it's more idiomatic than generate_gen.
The exl2 container's generate_gen function is now internal.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
The model card is a unified structure for sharing model params.
Rather than kwargs, use this instead.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
This is applied across containers. Doesn't make sense to put this method
in the backend.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
If a model is being unloaded, that means its being shut down and
no requests should be accepted from then on.
Also, remove model_is_loaded since we simply check if the container
is None now.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Maybe move this out of the class entirely, but for now, it makes
sense to encapsulate this logic.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Apparently the "mirostat" parameter has been updated by frontends
to pass a number. ExllamaV2 expects a boolean, but most pass a number
anyway, so just alias mirostat_mode and mirostat together.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
kwargs is pretty ugly when figuring out which arguments to use. The
base requests falls back to defaults anyways, so pass in the params
object as is.
However, since Python's typing isn't like TypeScript where types
can be transformed, the type hinting has a possiblity of None showing
up despite there always being a value for some params.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
The base sampler request already specifies the defaults, so don't
unwrap in this way.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Log the parameters passed into the generate gen function rather than
the generation settings to reduce complexity.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
This shouldn't even be an exposed option since changing it always
breaks inference with the model. Let the model's config.json handle
it.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
For the TP loader, GPU split cannot be an empty array. However,
defaulting the parameter to an empty array makes it easier to calculate
the device list. Therefore, cast an empty array to None using
falsy comparisons at load time.
Also add draft_gpu_split to the load request.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
Torch - 2.6.0
ExllamaV2 - 0.2.8
Flash-attn - 2.7.4.post1
Cuda wheels are now 12.4 instead of 12.1, feature names need to be
migrated over.
Signed-off-by: kingbri <8082010+kingbri1@users.noreply.github.com>
