Total runs:
671.3K
Run Growth:
386.1K
Growth Rate:
57.52%
Sponsored by
APIMart
AtomicChat / LFM2.5-24B-A2B-GGUF
Model's Last Updated: 10月 14 2026
text-generation
Introduction of LFM2.5-24B-A2B-GGUF
Model Details of LFM2.5-24B-A2B-GGUF
How to Run LFM2.5-24B-A2B Locally
Self-quantized from Liquid's original weights with our own importance matrix. The measurements behind this card are public.
- 24B mixture of experts with 2B active per token: the 15.4 GB build fits a 24 GB machine with room for context.
- We checked llama.cpp against Liquid's PyTorch code on the preview's first revision: llama.cpp's BF16 is closer to the exact fp32 model than PyTorch's own bf16 is.
- This model is unusually sensitive to numerical precision. Read the KLD below against its floor, not against other models'.
- Need it without refusals? LFM2.5-24B-A2B-Uncensored-GGUF has baked builds and a LoRA adapter for the quants here.
The builds
| Build | On disk | Mean KLD | Same top-1 |
|---|---|---|---|
BF16
|
48.0 GB | reference | reference |
AD-Q8_0
|
25.7 GB | 0.063 | 90.1% |
AD-Q6_K
|
20.9 GB | 0.075 | 89.1% |
AD-Q5_K_M
|
17.3 GB | 0.108 | 87.4% |
AD-Q4_K_M
|
15.4 GB | 0.126 | 85.5% |
AD-IQ4_XS
|
14.5 GB | 0.135 | 85.2% |
AD-IQ3_M
|
12.1 GB | 0.223 | 80.2% |
AD-IQ3_XXS
|
11.1 GB | 0.286 | 78.2% |
AD-Q4_K_M
is the one to take
on a 24 GB machine. With 32 GB,
AD-Q6_K
is
nearly as close to the original as
AD-Q8_0
at 5 GB less. Below 16 GB, the
3-bit builds lose noticeably more; on this model the step from 4 to 3 bits
nearly doubles the divergence.
Every number is against the BF16 model's own logits, measured on one machine. "Same top-1" is how often the build picks the same next token as the BF16 model.
Why the numbers look high
This model's output moves when the arithmetic changes, even when the weights do
not. We ran the same BF16 weights several ways, on the tokens and positions of
the reference above. This table was measured on the preview's first revision
(
c715ae25
); the current weights behave the same way at Q8_0 (0.063 against
0.064 then).
| Comparison | Mean KLD | Median KLD | Same top-1 |
|---|---|---|---|
| llama.cpp BF16, the same GPU twice | 0.000 | 0.000 | 100.0% |
| llama.cpp BF16, CPU against GPU | 0.019 | 0.002 | 95.2% |
| llama.cpp BF16 against exact fp32 (PyTorch) | 0.024 | 0.006 | 92.1% |
| Liquid's PyTorch bf16 against exact fp32 | 0.053 | 0.018 | 89.7% |
our
AD-Q8_0
against llama.cpp BF16
|
0.064 | 0.016 | 89.7% |
Rounding the activations to bf16 alone moves this model by 0.02-0.05 KLD, where
most models move by around a thousandth. A Q8_0 at 0.064 is therefore the floor
here, not a defect of the build. It also means llama.cpp runs this architecture
correctly: its BF16 sits closer to the exact model than Liquid's own bf16 code
path. Compare the builds with each other and with
AD-Q8_0
; comparing these KLD
values with those of other models says more about the models than the quants.
How these were made
-
BF16
from Liquid's original safetensors (revision
f772950c, the "agentic checkpoint" of 12 October), converted with llama.cpp980aef8c. The source is BF16, so nothing is requantized; the chat template is embedded. -
Importance matrix
on the BF16 model, never on a quantized proxy: 8,994
chunks of 512 tokens from our calibration corpus, rendered through this
model's own chat template, with
--parse-special. Agentic sessions and tool calls 24%, code 18%, multilingual text 20%, reasoning 13%, long documents 11%, structured data 4%, graphics code 4%, vocabulary sweep 6%. Every one of the 64 experts in all 38 MoE blocks saw at least 12.3k tokens, so no expert is quantized blind. -
Layout.
The rungs differ in the experts, which are nearly all of the
weights. Attention, the short-convolution mixers and the two dense blocks
stay at Q8_0 in every build, and the router is not quantized at all. The
embeddings are tied to the
output, so they stay at Q8_0 too (F16 in
AD-Q8_0andAD-Q6_K). Blocks 34 and 36-39 get a step up, chosen by where the importance matrix saw the most activation energy, not by a fixed fraction of the depth.
The importance matrix matters: on the first revision, the same
AD-Q4_K_M
layout without it measured 0.134 instead of 0.125.
Running them
llama-server -m LFM2.5-24B-A2B-AD-Q4_K_M.gguf -ngl 99 -c 32768 --jinja
Pass
--jinja
so the embedded chat template is used; without it llama.cpp falls
back to a generic one. Liquid's recommended sampling: temperature 0.1, top_k 50,
repetition penalty 1.05. Context goes up to 131,072 tokens. Needs a llama.cpp
build with
lfm2moe
and the LFM2.5 tokenizer, which upstream has.
In
Atomic Chat
:
search
AtomicChat/LFM2.5-24B-A2B-GGUF
, pick a build, hit
Use this model
.
Measurement
The reference is the BF16 model's logits over our held-out neutral set: 93
chunks of 4,096 tokens in 30 languages, BF16 PPL 15.67 ± 0.12. The calibration
corpus shares no 13-gram with it. One machine for everything: 1× H100 PCIe,
llama.cpp
980aef8c
. All logs, the importance matrix and the comparison above
are in the
metrics repository
.
Limitations
-
Built from the preview's revision
f772950c(12 October); the public release has the same weights, checked file by file. - Measured on general text only. Tool calling, long context and the thinking channel were not tested on the quantized builds.
- Not run yet on a Mac, in LM Studio or in Ollama.
License
LFM Open License v1.0, see
LICENSE
. Released by Liquid AI; quantized by Atomic Chat.
Runs of AtomicChat LFM2.5-24B-A2B-GGUF on huggingface.co
2.4K
Total runs
0
24-hour runs
263
3-day runs
2.4K
7-day runs
2.4K
30-day runs
More Information About LFM2.5-24B-A2B-GGUF huggingface.co Model
Url of LFM2.5-24B-A2B-GGUF
LFM2.5-24B-A2B-GGUF huggingface.co Url
Provider of LFM2.5-24B-A2B-GGUF huggingface.co
Other API from AtomicChat
Updated:10月 21 2026
Total runs:
428.5K
Run Growth:
421.1K
Growth Rate:
98.29%
Total runs:
230.5K
Run Growth:
-111.0K
Growth Rate:
-48.16%
Total runs:
160.6K
Run Growth:
154.5K
Growth Rate:
96.17%
Total runs:
55.7K
Run Growth:
-24.0K
Growth Rate:
-43.18%
Total runs:
48.6K
Run Growth:
-1.3K
Growth Rate:
-2.60%
Total runs:
44.8K
Run Growth:
44.8K
Growth Rate:
100.00%
Total runs:
42.5K
Run Growth:
-30.1K
Growth Rate:
-70.68%
Total runs:
42.2K
Run Growth:
-29.8K
Growth Rate:
-70.63%
Total runs:
42.2K
Run Growth:
-30.0K
Growth Rate:
-71.11%
Total runs:
22.4K
Run Growth:
-38.9K
Growth Rate:
-173.59%
Total runs:
19.8K
Run Growth:
17.8K
Growth Rate:
90.05%
Total runs:
17.3K
Run Growth:
13.3K
Growth Rate:
77.13%
Total runs:
16.1K
Run Growth:
8.8K
Growth Rate:
54.45%
Total runs:
10.8K
Run Growth:
10.4K
Growth Rate:
96.48%
Total runs:
10.6K
Run Growth:
6.4K
Growth Rate:
59.92%
Total runs:
10.4K
Run Growth:
8.1K
Growth Rate:
78.61%
Total runs:
9.9K
Run Growth:
2.3K
Growth Rate:
22.80%
Total runs:
8.0K
Run Growth:
7.6K
Growth Rate:
95.54%
Total runs:
8.0K
Run Growth:
7.5K
Growth Rate:
93.34%
Total runs:
7.4K
Run Growth:
5.1K
Growth Rate:
67.95%
Total runs:
6.3K
Run Growth:
4.3K
Growth Rate:
68.82%
Total runs:
6.3K
Run Growth:
1.5K
Growth Rate:
23.93%
Total runs:
5.9K
Run Growth:
157
Growth Rate:
2.67%
Total runs:
5.8K
Run Growth:
3.8K
Growth Rate:
66.13%
Total runs:
5.7K
Run Growth:
1.4K
Growth Rate:
24.57%
Total runs:
5.4K
Run Growth:
2.1K
Growth Rate:
38.30%
Total runs:
4.4K
Run Growth:
3.1K
Growth Rate:
71.07%
Total runs:
4.1K
Run Growth:
2.2K
Growth Rate:
54.00%
Total runs:
3.8K
Run Growth:
1.8K
Growth Rate:
47.61%
Total runs:
3.3K
Run Growth:
3.0K
Growth Rate:
90.13%
Total runs:
2.6K
Run Growth:
57
Growth Rate:
2.23%
Total runs:
2.5K
Run Growth:
2.5K
Growth Rate:
100.00%
Total runs:
1.9K
Run Growth:
-1.1K
Growth Rate:
-58.96%
Total runs:
1.8K
Run Growth:
-302
Growth Rate:
-17.15%
Total runs:
1.7K
Run Growth:
631
Growth Rate:
38.22%
Total runs:
1.6K
Run Growth:
1.6K
Growth Rate:
100.00%
Total runs:
1.6K
Run Growth:
1.6K
Growth Rate:
100.00%
Total runs:
1.6K
Run Growth:
-132
Growth Rate:
-8.41%
Total runs:
1.5K
Run Growth:
942
Growth Rate:
60.97%
Total runs:
1.3K
Run Growth:
-328
Growth Rate:
-24.96%
Total runs:
1.3K
Run Growth:
1.2K
Growth Rate:
95.36%
Total runs:
1.3K
Run Growth:
61
Growth Rate:
4.73%
Total runs:
1.3K
Run Growth:
1.1K
Growth Rate:
88.93%
Total runs:
1.3K
Run Growth:
-1.7K
Growth Rate:
-132.70%
Total runs:
1.2K
Run Growth:
1.2K
Growth Rate:
100.00%
Total runs:
1.0K
Run Growth:
1.0K
Growth Rate:
100.00%
Total runs:
909
Run Growth:
285
Growth Rate:
31.35%
Total runs:
777
Run Growth:
-30
Growth Rate:
-3.86%
Total runs:
631
Run Growth:
-287
Growth Rate:
-45.48%
Total runs:
624
Run Growth:
29
Growth Rate:
4.65%
Total runs:
617
Run Growth:
-447
Growth Rate:
-72.45%
Total runs:
602
Run Growth:
602
Growth Rate:
100.00%
Total runs:
525
Run Growth:
66
Growth Rate:
12.57%
Total runs:
517
Run Growth:
403
Growth Rate:
77.95%
Total runs:
502
Run Growth:
-107
Growth Rate:
-21.31%
Total runs:
494
Run Growth:
494
Growth Rate:
100.00%
Total runs:
458
Run Growth:
458
Growth Rate:
100.00%
Total runs:
429
Run Growth:
-426
Growth Rate:
-99.30%
Total runs:
422
Run Growth:
-48
Growth Rate:
-11.37%
Total runs:
390
Run Growth:
-38
Growth Rate:
-9.74%
Total runs:
357
Run Growth:
0
Growth Rate:
0.00%
Total runs:
334
Run Growth:
-763
Growth Rate:
-228.44%
Total runs:
318
Run Growth:
318
Growth Rate:
100.00%
Total runs:
295
Run Growth:
-399
Growth Rate:
-135.25%
Total runs:
286
Run Growth:
-43
Growth Rate:
-15.03%
Total runs:
267
Run Growth:
-106
Growth Rate:
-39.70%
Total runs:
237
Run Growth:
-841
Growth Rate:
-354.85%
Total runs:
211
Run Growth:
-46
Growth Rate:
-21.80%
Total runs:
199
Run Growth:
-343
Growth Rate:
-172.36%
Total runs:
193
Run Growth:
-69
Growth Rate:
-35.75%
Total runs:
193
Run Growth:
-22
Growth Rate:
-11.40%
Total runs:
192
Run Growth:
-428
Growth Rate:
-222.92%
Total runs:
190
Run Growth:
172
Growth Rate:
90.53%
Total runs:
188
Run Growth:
120
Growth Rate:
63.83%
Total runs:
187
Run Growth:
-44
Growth Rate:
-23.53%
Total runs:
184
Run Growth:
184
Growth Rate:
100.00%
Total runs:
174
Run Growth:
-264
Growth Rate:
-151.72%
Total runs:
174
Run Growth:
-140
Growth Rate:
-80.46%
Total runs:
161
Run Growth:
-34
Growth Rate:
-21.12%
Total runs:
158
Run Growth:
7
Growth Rate:
4.43%
Total runs:
154
Run Growth:
-61
Growth Rate:
-39.61%
Total runs:
152
Run Growth:
-160
Growth Rate:
-105.26%
Total runs:
142
Run Growth:
-137
Growth Rate:
-96.48%
Total runs:
117
Run Growth:
117
Growth Rate:
100.00%
Total runs:
117
Run Growth:
104
Growth Rate:
88.89%
Total runs:
112
Run Growth:
-38
Growth Rate:
-33.93%
Total runs:
105
Run Growth:
-287
Growth Rate:
-273.33%
Total runs:
94
Run Growth:
82
Growth Rate:
87.23%
Total runs:
92
Run Growth:
-77
Growth Rate:
-83.70%