GitLab

Deepseek V2 is a MoE model from Deepseek. Relevant variations
compared to other models:

- Grouped top-K in expert selection.
- mscale in yarn is calculated using the `mscale` and `mscale_all_dim`
  configuration options.
- `mscale_all_dim` is also used in scaling attention softmax.
- Permuting of the query/key representations before applying rotary
  embeddings.
- Some projections cannot be sharded (`q_a_proj`, `kv_a_proj_with_mqa`).
  So, we need weight loads that supports quantized weights. To this
  end `{Weights,WeightLoader}.get_weight` was added.
- The query/key head dimensionality differs from that of the value,
  so we need to pad during attention.
- Heads with size 192, needs an extension to our paged attention
  fork and we need to ensure that the KV cache is allocated with the
  correct size.
- Shared experts.

* Improve the handling of quantized weights

Handling of quantized weights was split between two mechanisms:

- For quantized checkpoints, we used the new weight loader
  infrastructure.
- For quantization while loading (EETQ, FP8, bitsandbytes) we
  instead relied on conditional in `get_linear`.

Weight loaders support context managers to selectively load
particular layers with different weight loaders, which is useful
for models like Idefics2 AWQ, which uses a quantized text model,
but unquantized vision and connector models. However, the context
manager would be overrided by `get_linear`, which string-checks
`quantizer`. Also, the context manager would not work with
EETQ, FP8, and bitsandbytes.

This change migrates all quantizers to the weight loader infrastructure.
This has several benefits:

- We can use context managers with all quantizers.
- All the implementation details move down to the quantizer layers,
  `get_linear` does not need to know how to handle quantizer linear
  layers.
- All quantizer weights are strongly typed, we don't pass around
  raw tensors.
- We don't have to pass around the `quantizer` string everywhere.

* Exclude non-MLP layers when using FP8 quantization with Llama

Fixes #2236.

Fixes #2036.

* feat: simple mistral lora integration tests

* fix: include args in docker launcher

* fix: disable cuda graphs with lora and warn

* fix: adjust docs and precommit issues

* fix: re update docs

Packing of asymmetric quantization is broken, all (q)zeros values
of `0` get reset to `1`, resulting in a loss of accuracy. So instead
use symmetric quantization. To be able to distinguish models with
symmetric and asymmetric quantization, a new config tensor `gptq_sym` is
added. If this tensor is not present, we assume `sym=False`.

Use FP8 GPTQ-Marlin kernels to enable FP8 support on CUDA GPUs
with compute capability >=8.0 and <8.9.
Co-authored-by: Florian Zimmermeister <flozi00.fz@gmail.com>

Quantized weights were loaded in the `Weights` class, but this was
getting quite unwieldy, where every higher level method to load weights
was a long conditional to cover all the different quantizers.

This change moves loading of quantized weights out of the `Weights`
class. This is done by defining a simple `WeightsLoader` interface
that is implemented by `Exl2WeightsLoader`, `GPTQWeightsLoader`,
and `MarlinWeightsLoader`. These implementations are in the quantizers'
respective modules. The `Weights` class provides the low-level load
operations (such as loading tensors or sharded tensors), but delegates
loads that need quantizer-specific weight processing to a loader. The
loaders still use the low-level functionality provided by `Weights`.

I initially tried making a hierarchy where a class like `GPTQWeights`
would inherit from `Weights`. But it is not very flexible (e.g. does
not work well with the new weight storage mock used in tests) and
the implicit indirections made the code harder to follow.

We wouldn't allocate any memory in multi-query (1 KV head). Fixes
Starcoder et al.

Fix number of KV heads

* Update idefics_causal_lm.py

Fix syntax issues

* fix dbrx & opt model prefix bug

* Consistently take `prefix` in model constructors

* Release test check fix

* Misc refactor-related fixes

* Refactor dead code.

* First working step.

* Remove a lot of duplicated code.

* More dead code.

* More cleanup.

* Fix Santacoder test.

* Fixing the simple tests.

* Fixing sharding.

* Fixes for VLM.

* Fixing santacoder (num_kv_heads hardcoded).

* Removing more dead code.

* Fixing `config.n_head`.

* Stopping earlier because of `<end_of_utterance>` in idefics2.

* Addresses comments.

* Removing the dead code.

* Fuse back mistral into FlashCausalLM.

* Finish removal.

* Fixing docs + causal_lm `batch_class`.

* Fixing docs + causal.lm.

* Add default to Gemma Causality.

* Default value for gemma/gemma2.

* Wrong default.

Adding "longrope" for phi-3

install triton because GPTQParams needs it.
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* Using flash decoding

Conditional flashdecoding.

Fix max_q.

Working kvcache

Working version with flash decoding.

Make it work for mistral.

Fix after rebase..

Less intrusive.

REvert changes in modeling.

Speedup flashdecoding.

HHachweew
Hack to make other models work.

Fixing non flash decoding llama path.

Router logic knows about page size.

Missing 2 models.

Missing cohere.

Fixing cohere flash decoding.

Revamped all this architecture.

Fix cohere.

Fixing falcon.

Enabling custom block size schedule.

Update router/src/infer.rs

Not sending preallocated output.

* Making it work on non flash decoding.

* Fix Cohere.

* Fix non decoding paths.

* Rebased.

* No need for cache_manager anymore.

* Update?

* "ipex" -> "cpu"

* These do not belong.

* Factoring cu_seqlen_qk for better abstracting over every model.

* Fixing non flash tests/imports.

* Changing return everywhere.

* Update mistral past.

* Fixing Mi{s,x}tral (non functional in Flash Decoding mode though).

* Fixup mistral clamping (had issues with cuda graphs).

* No need to recreate anything actually.

* refine get xpu free memory
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* enable qwen2 in xpu
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* enable gemma/gemma2/phi in intel platform
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

---------
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

https://github.com/huggingface/text-generation-inference/issues/2122

GPTQ-Marlin is currently the best-performing kernel for GPTQ models. So
let's use it by default if the kernels are installed, the GPU supports
it, and the kernels support the configuration.

For models generated by `text-generation-server quantize`, use
`sym=False`. This subcommand symmetric quantization since the beginning
and incorrectly reporting the model to be symmetric will use
GPTQ-Marlin (which does not support asymmetric quantization).

* Fixing gemma2.

* Adding new model.

Before this change, the number of reserved image tokens was not the
same as the number of images. Fixes #2029.

While at it, also remove all the image token handling duplication
in `prepare_input`.

This change adds support for 2:4 sparsity when using Marlin
quantization. The 2:4 kernel is used when:

* The quantizer is `marlin`;
* the quantizer checkpoint format is `marlin_24`.

Fixes #2098.

When the AWQ quantizer was used with a layer that uses a bias,
the bias tensor was not correctly passed/used. Instead, the
value `true`/`1.0` was added to the linear transformation.

Correctly pass through the bias when it is not `None`.

Fixes #2106.

* feat: first draft load multiple lora

* feat: load weights within layer and refactor lora pass

* fix: refactor and reduce lora math

* feat: baseline impl single request multi lora support

* feat: prefer lorax implementation and port loading logic

* fix: prefer adapter_data and refactors

* feat: perfer loraxs custom punica kernels and add mlp loras

* fix: adjust batch for bgmv

* fix: adjust adapter_segments logic when in batch

* fix: refactor and move changes to v3 proto

* fix: pass model_id for all flash causal lms

* fix: pass model_id for all causal and seq2seq lms

* fix: add model_id to model test

* feat: add lora support to mistral and refactors

* feat: prefer model id in request

* fix: include rust code for adapter id

* feat: bump launcher and add new lora docs

* feat: support base model generation and refactors

* fix: rename doc to retry ci build

* feat: support if vlm models

* fix: add adapter_data param and avoid missing layers

* fix: add adapter_data param to phi and neox

* fix: update all models forwards to include adapter_data

* fix: add model_id to IdeficsCausalLM

* Update lora.md

Fixed a typo

* Update lora.md

Fixing spam image

* fix: add lora kernel to dockerfile, support running without kernels and refactors

* fix: avoid dockerfile conflict

* fix: refactors and adjust flash llama lora logic

* fix: skip llama test due to CI issue (temp)

* fix: skip llama test CI (temp) 2

* fix: revert skips and prefer updated ci token for tests

* fix: refactors and helpful comments

* fix: add noop in TensorParallelAdapterRowLinear too

* fix: refactor and move shard_lora_weights logic

* fix: exit early if no adapter_data

---------
Co-authored-by: Derek <datavistics@gmail.com>