Skip to content

Conversation

@tdakhran
Copy link
Contributor

@tdakhran tdakhran commented Sep 23, 2025

HF link: LiquidAI/LFM2-2.6B.

Support for GGUF conversion and inference is added in #14620.

However, due to similar n_embd, it identifies as a 1.2B model. Fix the label by using n_ff to identify the model instead.

Output of llama-bench:

| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |

GGUFs will be uploaded to https://huggingface.co/LiquidAI/LFM2-2.6B-GGUF soon.

UPD: GGUFs are uploaded.

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in ggml-org#14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```
@tdakhran tdakhran requested a review from CISC as a code owner September 23, 2025 16:33
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
@CISC CISC merged commit 3a59971 into ggml-org:master Sep 24, 2025
1 check passed
@tdakhran tdakhran deleted the tarek/lfm2_2.6b branch September 24, 2025 11:46
pwilkin pushed a commit to pwilkin/llama.cpp that referenced this pull request Sep 25, 2025
* model : add label for LiquidAI LFM2-2.6B model

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in ggml-org#14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```

* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
struct pushed a commit to struct/llama.cpp that referenced this pull request Sep 26, 2025
* model : add label for LiquidAI LFM2-2.6B model

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in ggml-org#14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```

* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
yael-works pushed a commit to yael-works/llama.cpp that referenced this pull request Oct 15, 2025
* model : add label for LiquidAI LFM2-2.6B model

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in ggml-org#14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```

* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
pwilkin pushed a commit to pwilkin/llama.cpp that referenced this pull request Oct 23, 2025
* model : add label for LiquidAI LFM2-2.6B model

HF link: [LiquidAI/LFM2-2.6B](https://huggingface.co/LiquidAI/LFM2-2.6B).

Support for GGUF conversion and inference is added in ggml-org#14620.

However, due to similar `n_embd`, it identifies as a 1.2B model.
Fix the label by using `n_ff` to identify the model instead.

Output of `llama-bench`:
```
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| lfm2 1.2B F16                  |   2.18 GiB |     1.17 B | CPU        |      10 |           pp512 |        223.97 ± 5.32 |
| lfm2 2.6B F16                  |   4.79 GiB |     2.57 B | CPU        |      10 |           pp512 |         92.53 ± 4.14 |
| lfm2 350M F16                  | 676.25 MiB |   354.48 M | CPU        |      10 |           pp512 |       725.52 ± 11.70 |
| lfm2 700M F16                  |   1.38 GiB |   742.49 M | CPU        |      10 |           pp512 |       336.22 ± 12.93 |
```

* Update src/llama-model.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants