Laborator commited on
Commit
b125e41
·
verified ·
1 Parent(s): 6b5ae85

docs: replace auto-generated card with full MicroLens card (hackathon submission)

Browse files
Files changed (1) hide show
  1. README.md +130 -175
README.md CHANGED
@@ -1,210 +1,165 @@
1
  ---
2
- base_model: unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
 
3
  library_name: peft
4
- pipeline_tag: text-generation
5
  tags:
6
- - base_model:adapter:unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
7
- - lora
8
- - sft
9
- - transformers
10
- - trl
 
 
11
  - unsloth
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
  ---
13
 
14
- # Model Card for Model ID
15
 
16
- <!-- Provide a quick summary of what the model is/does. -->
17
 
 
18
 
 
19
 
20
- ## Model Details
 
 
 
 
 
 
 
 
21
 
22
- ### Model Description
23
 
24
- <!-- Provide a longer summary of what this model is. -->
25
 
 
 
 
 
26
 
 
27
 
28
- - **Developed by:** [More Information Needed]
29
- - **Funded by [optional]:** [More Information Needed]
30
- - **Shared by [optional]:** [More Information Needed]
31
- - **Model type:** [More Information Needed]
32
- - **Language(s) (NLP):** [More Information Needed]
33
- - **License:** [More Information Needed]
34
- - **Finetuned from model [optional]:** [More Information Needed]
35
 
36
- ### Model Sources [optional]
 
 
 
 
37
 
38
- <!-- Provide the basic links for the model. -->
 
 
 
 
 
 
 
39
 
40
- - **Repository:** [More Information Needed]
41
- - **Paper [optional]:** [More Information Needed]
42
- - **Demo [optional]:** [More Information Needed]
 
 
 
 
 
43
 
44
- ## Uses
45
 
46
- <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
 
 
47
 
48
- ### Direct Use
49
 
50
- <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
51
 
52
- [More Information Needed]
 
 
 
 
 
 
53
 
54
- ### Downstream Use [optional]
55
 
56
- <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
57
 
58
- [More Information Needed]
 
 
 
 
59
 
60
- ### Out-of-Scope Use
61
 
62
- <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
63
 
64
- [More Information Needed]
 
 
 
 
65
 
66
- ## Bias, Risks, and Limitations
67
 
68
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
69
 
70
- [More Information Needed]
 
 
 
 
 
71
 
72
- ### Recommendations
73
 
74
- <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75
 
76
- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
77
-
78
- ## How to Get Started with the Model
79
-
80
- Use the code below to get started with the model.
81
-
82
- [More Information Needed]
83
-
84
- ## Training Details
85
-
86
- ### Training Data
87
-
88
- <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
89
-
90
- [More Information Needed]
91
-
92
- ### Training Procedure
93
-
94
- <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
95
-
96
- #### Preprocessing [optional]
97
-
98
- [More Information Needed]
99
-
100
-
101
- #### Training Hyperparameters
102
-
103
- - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
104
-
105
- #### Speeds, Sizes, Times [optional]
106
-
107
- <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
108
-
109
- [More Information Needed]
110
-
111
- ## Evaluation
112
-
113
- <!-- This section describes the evaluation protocols and provides the results. -->
114
-
115
- ### Testing Data, Factors & Metrics
116
-
117
- #### Testing Data
118
-
119
- <!-- This should link to a Dataset Card if possible. -->
120
-
121
- [More Information Needed]
122
-
123
- #### Factors
124
-
125
- <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
126
-
127
- [More Information Needed]
128
-
129
- #### Metrics
130
-
131
- <!-- These are the evaluation metrics being used, ideally with a description of why. -->
132
-
133
- [More Information Needed]
134
-
135
- ### Results
136
-
137
- [More Information Needed]
138
-
139
- #### Summary
140
-
141
-
142
-
143
- ## Model Examination [optional]
144
-
145
- <!-- Relevant interpretability work for the model goes here -->
146
-
147
- [More Information Needed]
148
-
149
- ## Environmental Impact
150
-
151
- <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
152
-
153
- Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
154
-
155
- - **Hardware Type:** [More Information Needed]
156
- - **Hours used:** [More Information Needed]
157
- - **Cloud Provider:** [More Information Needed]
158
- - **Compute Region:** [More Information Needed]
159
- - **Carbon Emitted:** [More Information Needed]
160
-
161
- ## Technical Specifications [optional]
162
-
163
- ### Model Architecture and Objective
164
-
165
- [More Information Needed]
166
-
167
- ### Compute Infrastructure
168
-
169
- [More Information Needed]
170
-
171
- #### Hardware
172
-
173
- [More Information Needed]
174
-
175
- #### Software
176
-
177
- [More Information Needed]
178
-
179
- ## Citation [optional]
180
-
181
- <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
182
-
183
- **BibTeX:**
184
-
185
- [More Information Needed]
186
-
187
- **APA:**
188
-
189
- [More Information Needed]
190
-
191
- ## Glossary [optional]
192
-
193
- <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
194
-
195
- [More Information Needed]
196
-
197
- ## More Information [optional]
198
-
199
- [More Information Needed]
200
-
201
- ## Model Card Authors [optional]
202
-
203
- [More Information Needed]
204
-
205
- ## Model Card Contact
206
-
207
- [More Information Needed]
208
- ### Framework versions
209
-
210
- - PEFT 0.18.0
 
1
  ---
2
+ license: apache-2.0
3
+ base_model: unsloth/gemma-4-E2B-it
4
  library_name: peft
5
+ pipeline_tag: image-text-to-text
6
  tags:
7
+ - microscopy
8
+ - vision-language
9
+ - diatoms
10
+ - fungal-spores
11
+ - biology
12
+ - bioindicator
13
+ - gemma-4
14
  - unsloth
15
+ - qlora
16
+ - multimodal
17
+ - on-device
18
+ - offline
19
+ datasets:
20
+ - sergheibrinza/microlens-vqa-hackathon
21
+ - sergheibrinza/microlens-images-hackathon
22
+ language:
23
+ - en
24
+ - de
25
+ - fr
26
+ - es
27
+ - it
28
+ - pt
29
+ - ru
30
+ - zh
31
+ - ja
32
+ - ko
33
  ---
34
 
35
+ # MicroLens — Final
36
 
37
+ **A pocket-microscope expert.** Vision-language model that identifies microscopy specimens — diatoms and fungal spores across 95 genera — names the genus, and explains morphology, habitat, and identification cues. Built on Gemma 4 E2B, runs offline on a 4 GB Android, speaks 140+ languages out of the box.
38
 
39
+ Submission to the **Kaggle Gemma 4 Good Hackathon 2026**.
40
 
41
+ ## Links
42
 
43
+ | Resource | URL |
44
+ |---|---|
45
+ | Live web demo | https://huggingface.co/spaces/Laborator/microlens |
46
+ | Live Kaggle notebook (T4, 9 min) | https://www.kaggle.com/code/sergheibrinza/microlens-final |
47
+ | GitHub (source, APK, Modelfile) | https://github.com/SergheiBrinza/microlens |
48
+ | Training VQA dataset (75,491 pairs) | https://www.kaggle.com/datasets/sergheibrinza/microlens-vqa-hackathon |
49
+ | Training images (75,491 PNGs) | https://www.kaggle.com/datasets/sergheibrinza/microlens-images-hackathon |
50
+ | Ollama (3 GB GGUF) | `ollama run brinzaengineeringai/microlens-final` |
51
+ | Android APK | https://github.com/SergheiBrinza/microlens/releases |
52
 
53
+ ## What this model is
54
 
55
+ A 4-bit QLoRA fine-tune of `unsloth/gemma-4-E2B-it` that turns a generic vision-language model into a structured microscopy assistant. For any specimen image, MicroLens returns four sections:
56
 
57
+ - **Genus** (and species when it is sure)
58
+ - **Morphology** — shape, size, raphe, frustule
59
+ - **Habitat** — where this organism typically lives
60
+ - **Identification cues** — what to look for in the image
61
 
62
+ Covers **95 genera** across two categories: diatoms (the standard bioindicator behind the EU Water Framework Directive) and fungal spores.
63
 
64
+ ## Quick start (Python + Unsloth)
 
 
 
 
 
 
65
 
66
+ ```python
67
+ from unsloth import FastVisionModel
68
+ from peft import PeftModel
69
+ from PIL import Image
70
+ import torch
71
 
72
+ base, tokenizer = FastVisionModel.from_pretrained(
73
+ 'unsloth/gemma-4-E2B-it',
74
+ load_in_4bit=True,
75
+ use_gradient_checkpointing='unsloth',
76
+ max_seq_length=2048,
77
+ )
78
+ model = PeftModel.from_pretrained(base, 'Laborator/microlens-final')
79
+ FastVisionModel.for_inference(model)
80
 
81
+ img = Image.open('your_specimen.png').convert('RGB')
82
+ prompt = 'Identify the organism in this microscopy image and describe its morphology.'
83
+ msgs = [{'role':'user','content':[{'type':'image'},{'type':'text','text':prompt}]}]
84
+ text = tokenizer.apply_chat_template(msgs, add_generation_prompt=True)
85
+ inp = tokenizer(img, text, add_special_tokens=False, return_tensors='pt').to('cuda')
86
+ out = model.generate(**inp, max_new_tokens=200, do_sample=False)
87
+ print(tokenizer.decode(out[0][inp.input_ids.shape[-1]:], skip_special_tokens=True))
88
+ ```
89
 
90
+ ## Quick start (Ollama, on-device)
91
 
92
+ ```bash
93
+ ollama run brinzaengineeringai/microlens-final
94
+ ```
95
 
96
+ Pulls the 3 GB Q4_K_M GGUF and runs entirely on CPU or any consumer GPU.
97
 
98
+ ## Training summary
99
 
100
+ - **Base model:** `unsloth/gemma-4-E2B-it` (4.44 B parameters, ~2 B effective via Per-Layer Embeddings)
101
+ - **Method:** 4-bit QLoRA via Unsloth FastVisionModel, both vision tower and language tower trainable
102
+ - **Data:** 75,491 VQA pairs (67,121 train + 8,370 val), 95 genera, 2 categories
103
+ - **Schedule:** 2 epochs, 8,392 steps, lr 2e-4 cosine, batch 2×8=16, AdamW-8bit, bf16, seq 2048
104
+ - **Hardware:** 1× RTX 3090 Ti (24 GB), 14.7 hours wall-clock
105
+ - **Trainable params:** 29.9 M (0.58% of base), LoRA r=16, α=32
106
+ - **Final eval loss:** 0.0189 (smooth monotone decrease)
107
 
108
+ ## Evaluation results
109
 
110
+ Stratified 200-pair validation, 150 diatom + 50 fungal spore.
111
 
112
+ | Metric | Diatom (n=150) | Fungal spore (n=50) | Overall (n=200) |
113
+ |---|---|---|---|
114
+ | **Genus accuracy** (substring match) | 85.3% | **100%** | **89.0%** |
115
+ | **Category accuracy** | 100% | 100% | **100%** |
116
+ | **Format adherence** (morphology + habitat + cues) | 95.3% | 72.0% | **89.5%** |
117
 
118
+ Reproducible end to end on a free Kaggle T4 in 9 minutes — see the linked Kaggle notebook.
119
 
120
+ ## Training data — license-clean for commercial use
121
 
122
+ | Source | License | Pairs (train) |
123
+ |---|---|---|
124
+ | UDE Diatoms in the Wild 2024 (Zenodo 10410655) | CC0 | 39,389 |
125
+ | DIATLAS (Zenodo 16260887) | CC-BY 4.0 | 23,544 |
126
+ | TgFC — Tectona grandis fungal community (figshare 28855910) | CC-BY 4.0 | 4,188 |
127
 
128
+ Top-30 genera have hand-curated knowledge-base answers from AlgaeBase, WoRMS, ITIS. ZooScanNet (CC-BY-NC) and ZooLake (license not verifiable) were intentionally excluded so this release is clean for commercial use end to end.
129
 
130
+ ## Honest limits
131
 
132
+ - Trained on stained light-microscopy at 384×384. SEM and fluorescence are out of distribution.
133
+ - Only 95 genera (diatoms + fungal spores). Plankton, pollen and soil microbes were in earlier internal builds but got dropped to keep licensing clean.
134
+ - Long-tail genera produce shorter answers. The curated knowledge base only covers the top 30.
135
+ - Confidence is expressed in words ("looks like X but the asymmetry suggests Y"), not calibrated probabilities. Good for an explainable assistant, bad for automated decisions.
136
+ - No held-out test split. The 8,370 val pairs do double duty for per-step and final eval. A future release will fix that.
137
+ - **Research artefact — not a medical device. Not for clinical, diagnostic, or regulatory use.**
138
 
139
+ ## License & attribution
140
 
141
+ Apache 2.0 — matches base Gemma 4 license. Please credit *Serghei Brinza — MicroLens, Vienna 2026*.
142
+
143
+ ## Citation
144
+
145
+ If you use MicroLens in research, please cite:
146
+
147
+ ```bibtex
148
+ @misc{brinza2026microlens,
149
+ author = {Serghei Brinza},
150
+ title = {MicroLens: A Pocket-Microscope Expert via Gemma 4 E2B},
151
+ year = 2026,
152
+ publisher = {Hugging Face},
153
+ howpublished = {\url{https://huggingface.co/Laborator/microlens-final}},
154
+ note = {Kaggle Gemma 4 Good Hackathon 2026 submission}
155
+ }
156
+ ```
157
+
158
+ Also cite the upstream:
159
 
160
+ - Gemma 4 (Google DeepMind)
161
+ - Unsloth (Daniel & Michael Han) — https://github.com/unslothai/unsloth
162
+ - AlgaeBase, WoRMS, ITIS — taxonomic knowledge bases
163
+ - UDE Diatoms in the Wild 2024 (Zenodo 10410655)
164
+ - DIATLAS (Zenodo 16260887)
165
+ - TgFC (figshare 28855910)