Feature Extraction
Transformers
Safetensors
lfm2
fill-mask
encoder-only
multimodal
image-text-retrieval
image-text-matching
siglip2
lfm2.5
gptq
custom_code
compressed-tensors
Instructions to use konic-labs/LFM2.5-multimodal-encoder-230M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use konic-labs/LFM2.5-multimodal-encoder-230M with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="konic-labs/LFM2.5-multimodal-encoder-230M", trust_remote_code=True)# Load model directly from transformers import AutoTokenizer, AutoModelForMaskedLM tokenizer = AutoTokenizer.from_pretrained("konic-labs/LFM2.5-multimodal-encoder-230M", trust_remote_code=True) model = AutoModelForMaskedLM.from_pretrained("konic-labs/LFM2.5-multimodal-encoder-230M", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
| { | |
| "selected_artifact": "/data/lfm2.5_encoder_lab/outputs/monet-100k/final-gptq-int4-clean", | |
| "selection_role": "publication-oriented clean image-only encoder branch", | |
| "training_correction": "image branch never receives candidate caption during contrastive training or retrieval evaluation", | |
| "dataset": "/data/datasets/monet-100k", | |
| "external_eval": "/data/datasets/deep-eval-cc3m-2k", | |
| "compression": "GPTQ INT4 W4A16 with fused calibration", | |
| "size_MB": 370.4, | |
| "clean_monet_bf16": { | |
| "pair_count": 12500, | |
| "image_to_text": { | |
| "recall_at_1": 0.11935999989509583, | |
| "recall_at_5": 0.3088800013065338, | |
| "recall_at_10": 0.4241600036621094, | |
| "mean_rank": 58.65480041503906, | |
| "median_rank": 16.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.11336000263690948, | |
| 0.12439999729394913 | |
| ], | |
| "recall_at_5": [ | |
| 0.2990399897098541, | |
| 0.3176800012588501 | |
| ], | |
| "recall_at_10": [ | |
| 0.4156000018119812, | |
| 0.432559996843338 | |
| ], | |
| "mean_rank": [ | |
| 56.50775909423828, | |
| 61.287200927734375 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "text_to_image": { | |
| "recall_at_1": 0.15783999860286713, | |
| "recall_at_5": 0.3598400056362152, | |
| "recall_at_10": 0.4731200039386749, | |
| "mean_rank": 50.04256057739258, | |
| "median_rank": 12.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.1515199989080429, | |
| 0.1632000058889389 | |
| ], | |
| "recall_at_5": [ | |
| 0.3503200113773346, | |
| 0.36711999773979187 | |
| ], | |
| "recall_at_10": [ | |
| 0.4652799963951111, | |
| 0.4806399941444397 | |
| ], | |
| "mean_rank": [ | |
| 47.80712127685547, | |
| 52.608638763427734 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "finite_hidden": true | |
| }, | |
| "clean_monet_gptq": { | |
| "pair_count": 12500, | |
| "image_to_text": { | |
| "recall_at_1": 0.10927999764680862, | |
| "recall_at_5": 0.2959200143814087, | |
| "recall_at_10": 0.4089600145816803, | |
| "mean_rank": 63.38008117675781, | |
| "median_rank": 17.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.10367999970912933, | |
| 0.11488000303506851 | |
| ], | |
| "recall_at_5": [ | |
| 0.28672000765800476, | |
| 0.30616000294685364 | |
| ], | |
| "recall_at_10": [ | |
| 0.39952000975608826, | |
| 0.4176799952983856 | |
| ], | |
| "mean_rank": [ | |
| 61.031761169433594, | |
| 65.88712310791016 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "text_to_image": { | |
| "recall_at_1": 0.1459999978542328, | |
| "recall_at_5": 0.341839998960495, | |
| "recall_at_10": 0.45903998613357544, | |
| "mean_rank": 53.78031921386719, | |
| "median_rank": 13.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.13991999626159668, | |
| 0.15112000703811646 | |
| ], | |
| "recall_at_5": [ | |
| 0.3343999981880188, | |
| 0.34887999296188354 | |
| ], | |
| "recall_at_10": [ | |
| 0.44944000244140625, | |
| 0.46775999665260315 | |
| ], | |
| "mean_rank": [ | |
| 51.349281311035156, | |
| 56.50944137573242 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "finite_hidden": true | |
| }, | |
| "clean_cc3m_gptq": { | |
| "pair_count": 2000, | |
| "image_to_text": { | |
| "recall_at_1": 0.08649999648332596, | |
| "recall_at_5": 0.21899999678134918, | |
| "recall_at_10": 0.3059999942779541, | |
| "mean_rank": 137.33999633789062, | |
| "median_rank": 35.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.07349999994039536, | |
| 0.09950000047683716 | |
| ], | |
| "recall_at_5": [ | |
| 0.20100000500679016, | |
| 0.2409999966621399 | |
| ], | |
| "recall_at_10": [ | |
| 0.2854999899864197, | |
| 0.3269999921321869 | |
| ], | |
| "mean_rank": [ | |
| 127.39800262451172, | |
| 148.1855010986328 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "text_to_image": { | |
| "recall_at_1": 0.1145000010728836, | |
| "recall_at_5": 0.28349998593330383, | |
| "recall_at_10": 0.3865000009536743, | |
| "mean_rank": 112.76300048828125, | |
| "median_rank": 19.0, | |
| "bootstrap_95ci": { | |
| "recall_at_1": [ | |
| 0.10100000351667404, | |
| 0.13099999725818634 | |
| ], | |
| "recall_at_5": [ | |
| 0.26499998569488525, | |
| 0.30300000309944153 | |
| ], | |
| "recall_at_10": [ | |
| 0.367000013589859, | |
| 0.4115000069141388 | |
| ], | |
| "mean_rank": [ | |
| 102.36699676513672, | |
| 124.84700012207031 | |
| ], | |
| "resamples": 200 | |
| } | |
| }, | |
| "finite_hidden": true | |
| }, | |
| "matching": { | |
| "monet_bf16": { | |
| "cyclic_negative": { | |
| "pairs": 12500, | |
| "accuracy_at_zero": 0.9342, | |
| "positive_beats_negative": 0.981440007686615, | |
| "positive_mean": 2.9899163246154785, | |
| "negative_mean": -6.390422344207764, | |
| "margin_mean": 9.380338668823242, | |
| "auroc": 0.9746800661087036, | |
| "auprc": 0.9634296875 | |
| }, | |
| "hard_negative": { | |
| "pairs": 12500, | |
| "accuracy_at_zero": 0.50116, | |
| "positive_beats_negative": 0.48607999086380005, | |
| "positive_mean": 2.9899163246154785, | |
| "negative_mean": 2.9905307292938232, | |
| "margin_mean": -0.0006143388454802334, | |
| "auroc": 0.49737921357154846, | |
| "auprc": 0.497440703125 | |
| }, | |
| "zero_visual_positive": { | |
| "positive_mean": -0.28617942333221436, | |
| "positive_minus_trained_mean": 3.2760958671569824 | |
| } | |
| }, | |
| "monet_gptq": { | |
| "cyclic_negative": { | |
| "pairs": 12500, | |
| "accuracy_at_zero": 0.93128, | |
| "positive_beats_negative": 0.9807999730110168, | |
| "positive_mean": 2.9612369537353516, | |
| "negative_mean": -6.149127960205078, | |
| "margin_mean": 9.110363960266113, | |
| "auroc": 0.9733310341835022, | |
| "auprc": 0.96191859375 | |
| }, | |
| "hard_negative": { | |
| "pairs": 12500, | |
| "accuracy_at_zero": 0.50064, | |
| "positive_beats_negative": 0.4914399981498718, | |
| "positive_mean": 2.9612369537353516, | |
| "negative_mean": 2.967132568359375, | |
| "margin_mean": -0.005896187853068113, | |
| "auroc": 0.49611741304397583, | |
| "auprc": 0.49736484375 | |
| }, | |
| "zero_visual_positive": { | |
| "positive_mean": -0.13991403579711914, | |
| "positive_minus_trained_mean": 3.1011509895324707 | |
| } | |
| }, | |
| "cc3m_gptq": { | |
| "cyclic_negative": { | |
| "pairs": 2000, | |
| "accuracy_at_zero": 0.7265, | |
| "positive_beats_negative": 0.8690000176429749, | |
| "positive_mean": 1.1929634809494019, | |
| "negative_mean": -1.334358811378479, | |
| "margin_mean": 2.527322292327881, | |
| "auroc": 0.8065452575683594, | |
| "auprc": 0.7932015380859375 | |
| }, | |
| "hard_negative": { | |
| "pairs": 2000, | |
| "accuracy_at_zero": 0.5185, | |
| "positive_beats_negative": 0.5665000081062317, | |
| "positive_mean": 1.1929634809494019, | |
| "negative_mean": 1.0014755725860596, | |
| "margin_mean": 0.19148792326450348, | |
| "auroc": 0.525624692440033, | |
| "auprc": 0.5216671752929688 | |
| }, | |
| "zero_visual_positive": { | |
| "positive_mean": 0.3554673492908478, | |
| "positive_minus_trained_mean": 0.8374961614608765 | |
| } | |
| } | |
| }, | |
| "fidelity": { | |
| "hidden_cosine_mean": 0.9427031874656677, | |
| "logits_cosine_mean": 0.965385913848877, | |
| "logits_top1_agreement": 0.8484848737716675 | |
| }, | |
| "reports": "/data/lfm2.5_encoder_lab/reports/deep_eval", | |
| "caveats": [ | |
| "hard-negative AUROC is near chance", | |
| "masked-token prediction does not improve over zero-image controls", | |
| "external subset is bounded CC3M and source terms require review", | |
| "no generation/decoder is included" | |
| ] | |
| } | |