Skip to content

use --use_sdpa_with_kv_cache for 1B/3B bf16 #5861

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 3 commits into from
Closed

use --use_sdpa_with_kv_cache for 1B/3B bf16 #5861

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
95c889a
use --use_sdpa_with_kv_cache for 1B/3B bf16
helunwencser Oct 3, 2024
00231ed
Update base for Update on "use --use_sdpa_with_kv_cache for 1B/3B bf16"
helunwencser Oct 4, 2024
ac207db
Update on "use --use_sdpa_with_kv_cache for 1B/3B bf16"
helunwencser Oct 4, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion examples/demo-apps/android/LlamaDemo/docs/delegates/xnnpack_README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ We have supported BFloat16 as a data type on the XNNPACK backend for Llama 3.2 1
* Export Llama model and generate .pte file as below:

```
python -m examples.models.llama2.export_llama --checkpoint <checkpoint.pth> --params <params.json> -kv -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2.pte"
python -m examples.models.llama2.export_llama --checkpoint <checkpoint.pth> --params <params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2.pte"
```

* Rename tokenizer for Llama 3.2 with command: `mv tokenizer.model tokenizer.bin`. We are updating the demo app to support tokenizer in original format directly.
Expand Down
2 changes: 1 addition & 1 deletion examples/demo-apps/apple_ios/LLaMA/docs/delegates/xnnpack_README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@ We have supported BFloat16 as a data type on the XNNPACK backend for Llama 3.2 1
* Export Llama model and generate .pte file as below:

```
python -m examples.models.llama2.export_llama --checkpoint <checkpoint.pth> --params <params.json> -kv -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2.pte"
python -m examples.models.llama2.export_llama --checkpoint <checkpoint.pth> --params <params.json> -kv --use_sdpa_with_kv_cache -X -d bf16 --metadata '{"get_bos_id":128000, "get_eos_ids":[128009, 128001]}' --output_name="llama3_2.pte"
```

For more detail using Llama 3.2 lightweight models including prompt template, please go to our official [website](https://www.llama.com/docs/model-cards-and-prompt-formats/llama3_2#-llama-3.2-lightweight-models-(1b/3b)-).
Expand Down
Binary file modified examples/models/llama2/Android3_2_1B_bf16.gif
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
12 changes: 7 additions & 5 deletions examples/models/llama2/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,10 +85,10 @@ We have verified running Llama 2 7B [mobile applications](#step-6-build-mobile-a
### Llama 3.2 1B and 3B
Llama 3.2 1B and 3B performance was measured on the OnePlus 12 device. The performance measurement is expressed in terms of tokens per second using an [adb binary-based approach](#step-5-run-benchmark-on) for generating 128 tokens.

|Model | 4bit(*) via SpinQuant
|--------| ---------------
|1B | 53.41 tokens/second |
|3B | 22.98 tokens/second |
|Model | bf16 | 4bit(*) via SpinQuant
|--------| ---------------------- | ---------------
|1B | 19.4 tokens/second | 53.41 tokens/second |
|3B | 7.76 tokens/second | 22.98 tokens/second |

(*) With SpinQuant, we currently quantize 4-bit groupwise (with groupsize 32) weight, 8bit dynamic activation of all the linear layers of the model, except embedding and output layers. The embedding and output layers are quantized as 8-bit per-channel weight and 8-bit dynamic activation.

Expand Down Expand Up @@ -142,7 +142,9 @@ LLAMA_PARAMS=path/to/params.json
python -m examples.models.llama2.export_llama \
--checkpoint "${LLAMA_CHECKPOINT:?}" \
--params "${LLAMA_PARAMS:?}" \
-kv -X \
-kv \
--use_sdpa_with_kv_cache \
-X \
-d bf16 \
--metadata '{"append_eos_to_prompt": 0, "get_bos_id":128000, "get_eos_ids":[128009, 128001], "get_n_bos": 0, "get_n_eos": 0}' \
--output_name="llama3_2.pte"
Expand Down
Loading