Added Infra in QEfficient for execution of swiftkv models #314
+543
−3
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
quic-amitraj
requested changes
Mar 17, 2025
QEfficient/transformers/models/llama_swiftkv/modeling_llama_swiftkv.py
Outdated
Show resolved
Hide resolved
QEfficient/transformers/models/llama_swiftkv/modeling_llama_swiftkv.py
Outdated
Show resolved
Hide resolved
…e is not present at hugging face and checkpoint like swiftkv Signed-off-by: Hem Agnihotri <[email protected]>
Signed-off-by: Hem Agnihotri <[email protected]>
Signed-off-by: Onkar Chougule <[email protected]>
Signed-off-by: Hem Agnihotri <[email protected]>
Signed-off-by: Amit Raj <[email protected]>
quic-amitraj
requested changes
Apr 2, 2025
QEfficient/transformers/models/llama_swiftkv/modeling_llama_swiftkv.py
Outdated
Show resolved
Hide resolved
Signed-off-by: Amit Raj <[email protected]>
quic-rishinr
requested changes
Apr 7, 2025
| super().__init__() | ||
| self.hidden_size = config.hidden_size | ||
| self.attention_dropout = config.attention_dropout | ||
| self.hidden_size = config.hidden_size |
There was a problem hiding this comment.
nit: please remove the redundant self.hidden_size initialization
| def forward( | ||
| self, | ||
| hidden_states: torch.Tensor, | ||
| position_ids, |
There was a problem hiding this comment.
nit: please add the hint type for position_ids and attention_mask for consistency. (Applicable for classes below as well)
| "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class " | ||
| "with a layer index." | ||
| ) | ||
| kv_seq_len = past_key_value.get_usable_length(kv_seq_len, self.layer_idx) |
There was a problem hiding this comment.
is it the right usage? According to the hint type past_key_value is Tuple[torch.Tensor, torch.Tensor]
| ) | ||
| kv_seq_len = past_key_value.get_usable_length(kv_seq_len, self.layer_idx) | ||
| cache_kwargs = {"position_ids": position_ids, "batch_index": batch_index} | ||
| key_states, value_states = past_key_value.read_only(self.layer_idx, cache_kwargs=cache_kwargs) |
| key_states = repeat_kv(key_states, self.num_key_value_groups) | ||
| value_states = repeat_kv(value_states, self.num_key_value_groups) | ||
|
|
||
| attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim) |
There was a problem hiding this comment.
Is these attention implementation still applicable with transformer 4.50?
| self, | ||
| hidden_states: torch.Tensor, | ||
| position_ids, | ||
| past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, |
There was a problem hiding this comment.
Ensure the naming consistency of the past_key_value parameter across all classes. In some classes, it is referred to as past_key_value, while in others, it is named past_key_values
| @@ -36,6 +36,58 @@ class QEffDynamicCache(DynamicCache): | |||
|
|
|||
| """ | |||
|
|
|||
| def write_only(self, key_states, value_states, layer_idx, cache_kwargs): | |||
| # Update the cache | |||
There was a problem hiding this comment.
please add doc string for both methods
| @@ -33,6 +33,7 @@ | |||
| | **Phi3ForCausalLM** | Phi-3, Phi-3.5 | [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) | ✔️ | | |||
| | **QwenForCausalLM** | DeepSeek-R1-Distill-Qwen | [DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) | ✔️ | | |||
| | | Qwen2, Qwen2.5 | [Qwen/Qwen2-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2-1.5B-Instruct) | ✔️ | | |||
| | **LlamaSwiftKVForCausalLM** | swiftkv | [Snowflake/Llama-3.1-SwiftKV-8B-Instruct](https://huggingface.co/Snowflake/Llama-3.1-SwiftKV-8B-Instruct) | ✔️ | | |||
There was a problem hiding this comment.
please update the feature support under docs/source/quick_start.md Supported Features
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Added Infra in QEfficient for execution of models whose modelling file is not present at hugging face and checkpoint like swiftkv