ducklingcodehouse commited on
Commit
9f93060
·
verified ·
0 Parent(s):

Super-squash: keep only latest state and drop old LFS blobs

Browse files
.gitattributes ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Finnish-NLP/Ahma-3B-Instruct
3
+ tags:
4
+ - transformers
5
+ - unsloth
6
+ - llama
7
+ - peft
8
+ - finnish
9
+ - medical
10
+ - dental
11
+ - healthcare
12
+ - research-only
13
+ license: apache-2.0
14
+ language:
15
+ - fi
16
+ library_name: peft
17
+ ---
18
+
19
+ # Finnish-DentalQA-lora
20
+
21
+ **Finnish-language conversational AI assistant specialized for dental medicine.** This is a fine-tuned large language model (LLM) that generates structured text responses to dental consultation queries, designed to simulate professional dentist-to-dentist consultations with clinical reasoning and recommendations.
22
+
23
+ **RESEARCH USE ONLY - NOT FOR MEDICAL DECISIONS**
24
+
25
+ - **Developed by:** [Heikki Saxén](https://fi.linkedin.com/in/heikkisaxen) / [Duckling Codehouse Oy](https://www.linkedin.com/company/duckling-codehouse-oy/) ([ducklingcodehouse](https://huggingface.co/ducklingcodehouse))
26
+ - **Supported by:** [Entteri Professional Software Oy](https://www.entteri.com/) (makers of AssisDent dental PMS)
27
+ - **License:** apache-2.0
28
+ - **Base Model:** Finnish-NLP/Ahma-3B-Instruct (by Aapo Tanskanen & Rasmus Toivanen)
29
+ - **Context Length:** 2048 tokens
30
+ - **Release Date:** August 20, 2025
31
+
32
+ LoRA adapter for Finnish dental medicine consultations between healthcare professionals. Trained on 30,908 synthetic dental conversations (80% expert clinical cases, 20% concept explanations) covering a broad range of different scenarios. Generated using pipeline with GPT-4.1.
33
+
34
+ **Research Focus:** This model demonstrates domain-specific fine-tuning with low computational resources. The goal is to explore how specialized models can be trained and deployed on consumer hardware (including personal GPUs) rather than requiring high-end infrastructure.
35
+
36
+ **System prompt recommendation:** This model was trained with a specific system prompt. For best results, we recommend using the same prompt format shown in the examples below.
37
+
38
+ **Response Format:** The model is trained to structure all clinical responses in three sections: "### Tausta" (Background), "### Arvio" (Assessment), "### Suositus" (Recommendation).
39
+
40
+ **Context Limit:** 2048 tokens total (input + output). Long conversations may need trimming to fit within this limit.
41
+
42
+ This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's Transformers library.
43
+
44
+ [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
45
+
46
+ ## Model Comparison
47
+
48
+ **LoRA Version**: Separate adapter files, requires base model
49
+
50
+ **Merged Version**: LoRA adapters merged into standalone model, no additional files needed
51
+
52
+ ## Installation
53
+
54
+ ```bash
55
+ pip install transformers torch peft accelerate
56
+ ```
57
+
58
+ ## Loading the Model
59
+
60
+ ```python
61
+ from transformers import AutoTokenizer, AutoModelForCausalLM
62
+ from peft import PeftModel
63
+ import torch
64
+
65
+ # Load base + LoRA
66
+ base_model = AutoModelForCausalLM.from_pretrained(
67
+ "Finnish-NLP/Ahma-3B-Instruct",
68
+ torch_dtype=torch.float16, # Recommended: FP16 for base model
69
+ device_map="auto"
70
+ )
71
+ model = PeftModel.from_pretrained(base_model, "ducklingcodehouse/Finnish-DentalQA-lora")
72
+
73
+ model = model.merge_and_unload()
74
+
75
+ # Load tokenizer from adapter model
76
+ tokenizer = AutoTokenizer.from_pretrained("ducklingcodehouse/Finnish-DentalQA-lora")
77
+ ```
78
+
79
+ ## Generate Response
80
+
81
+ ```python
82
+ # Use exact system prompt from training
83
+ system_prompt = """Olet kokenut suomalainen hammaslääkäri. Vastaat ammattimaisesti kollegojesi
84
+ kysymyksiin käyttäen oikeaa hammaslääketieteellistä terminologiaa ja viittaat Käypä hoito
85
+ -suosituksiin kun relevanttia."""
86
+
87
+ user_question = """87-vuotias nainen, jolla on vaikea muistisairaus, asuu palvelutalossa ja on
88
+ kotihoidon piirissä. Potilaalla on hampaattomuus, yläleuan kokoproteesi ja alaleuassa 6
89
+ luonnollista hammasta. Lääkityksenä on mm. rivaroksabaani eteisvärinään, donepetsiili,
90
+ metoprololi, kalsium+D-vitamiini, sekä useita muita lääkkeitä (yhteensä 9). Suun limakalvot
91
+ ovat kuivat, kielessä fissuroitumista, ja proteesin alla lievää punoitusta. HbA1c 46 mmol/mol,
92
+ BMI 22. Kotihoito raportoi potilaan nielevän huonosti ja syljen erityksen olevan vähäistä.
93
+ Karies- ja parodontiittiriski arvioidaan korkeaksi. Miten suosittelisit suun hoidon
94
+ järjestämistä ja priorisointia tässä tilanteessa?"""
95
+
96
+ messages = [
97
+ {"role": "system", "content": system_prompt},
98
+ {"role": "user", "content": user_question}
99
+ ]
100
+
101
+ inputs = tokenizer.apply_chat_template(
102
+ messages,
103
+ tokenize=True,
104
+ add_generation_prompt=True,
105
+ return_tensors="pt"
106
+ ).to(model.device)
107
+
108
+ # Recommended settings
109
+ with torch.no_grad():
110
+ outputs = model.generate(
111
+ inputs,
112
+ max_new_tokens=600,
113
+ temperature=0.1,
114
+ do_sample=True,
115
+ top_p=0.9,
116
+ repetition_penalty=1.2,
117
+ pad_token_id=tokenizer.pad_token_id
118
+ )
119
+
120
+ response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
121
+ print(response)
122
+
123
+ # Expected example response:
124
+ # ### Tausta
125
+ # Potilas on iäkäs, monisairas ja käyttää antikoagulanttihoitoa (rivaroksabaani), mikä
126
+ # lisää vuotoriskiä toimenpiteissä. Kuiva suu liittyy sekä lääkitykseen että ikääntymiseen.
127
+ # Hampaattomuus ja vähäinen syljeneritys altistavat limakalvovaurioille ja infektioille.
128
+ # Nielemisvaikeudet ja korkea karies- ja parodontiittiriski korostavat ehkäisevien
129
+ # toimenpiteiden merkitystä.
130
+ #
131
+ # ### Arvio
132
+ # Kliinisesti potilaalla on merkittävä hyposalivaatio (stimuloitu sylki <0,7 ml/min),
133
+ # joka yhdessä polyfarmasian kanssa lisää riskiä limakalvosairauksille ja infektioille.
134
+ # Proteesin alla oleva punoitus voi viitata proteesi-stomatiittiin tai kandidaasiin.
135
+ # Rivaroksabaanin vuoksi invasiivisia toimenpiteitä tulee välttää ellei välttämätöntä.
136
+ # Kotihoidon resurssit ja potilaan toimintakyky rajoittavat laajojen hoitojen toteutusta.
137
+ #
138
+ # ### Suositus
139
+ # Painopiste on limakalvojen kosteutuksessa (esim. keinotekoiset syljentuotteet) ja
140
+ # proteesin puhtauden ylläpidossa. Kotihoidolle annetaan selkeät ohjeet päivittäisestä
141
+ # proteesin puhdistuksesta ja limakalvojen tarkkailusta. Syljen eritystä voidaan tukea
142
+ # säännöllisellä nesteytyksellä ja tarvittaessa ksylitolituotteilla. Lääkärin kanssa
143
+ # arvioidaan mahdollisuudet vähentää kuivan suun aiheuttajia. Säännöllinen suun tarkastus
144
+ # (3–6 kk välein) on suositeltavaa. Käypä hoito -suositukset: "Lyhentyneen hammaskaaren
145
+ # hoito", "Karies (hallinta)" ja "Parodontiitti".
146
+ ```
147
+
148
+ ## Example
149
+
150
+ **User:** 27-vuotias mies, jolla on voimakas kipu ja turvotus alaleuan oikealla puolella. Viisaudenhammas 48 on puoliksi puhkeamassa, ympärillä runsasta ienturvotusta ja märkäkertymä. Potilaalla on kuumetta 38,3°C ja suun avausrajoitus 22 mm. Mitä ensihoitoa suosittelet terveyskeskuksessa?
151
+
152
+ **Assistant:**
153
+ ### Tausta
154
+ Potilas esittäytyy tyypillisin akuutin perikoroniitin oirein: kuume, paikallinen turvotus sekä trismus (avausrajoitus 22 mm). Puoliksi puhjenneen viisaudenhampaan (48) ympärillä oleva runsas ienturvotus ja märkäkertymä viittaavat paikalliseen infektioon, joka voi levitä syvempiin kudoksiin. Suun avausrajoitus (trismus) lisää riskiä infektion leviämiselle esimerkiksi submandibulaaritilaan.
155
+
156
+ ### Arvio
157
+ Kliininen kuva sopii akuuttiin perikoroniittiin komplisoituneena paikallisella absessilla tai flegmoonalla. Kuume ja yleisoireet täyttävät mikrobilääkehoidon aloituskriteerit Käypä hoito -suositusten mukaan. Trismus vaikeuttaa paikallista puhdistusta ja altistaa infektion leviämiselle. Märkäkertymän dreneeraus on ensisijaista, mutta myös antibioottihoito on tarpeen.
158
+
159
+ ### Suositus
160
+ Aloita amoksisilliini 500 mg x 3 po ja metronidatsoli 500 mg x 2 po suun kautta 7 päivän ajan. Kipulääkitys parasetamolilla ja/tai tulehduskipulääkkeellä. Huuhtele suu klooriheksidiinillä 1–2 kertaa päivässä. Mikäli potilaan yleistila heikkenee tai ilmaantuu nielemisvaikeuksia/hengitystieoireita, päivystyslähete erikoissairaanhoitoon on aiheellinen. Seuraa tilannetta tiiviisti; mikäli avausrajoitus pahenee tai yleisvointi huononee, sairaalahoito voi olla tarpeen.
161
+
162
+ **Follow-up:** Potilas palaa seuraavana päivänä. Kuume on noussut 39,2 °C:een, suun avaus onnistuu enää 10 mm, alaleuan oikea puoli on selvästi turvonnut ja nieleminen on kivuliasta. Hengitys on hieman rohisevaa. Mitä toimenpiteitä suosittelet tässä vaiheessa?
163
+
164
+ **Assistant:**
165
+ ### Tausta
166
+ Potilaan tila on selvästi pahentunut: kuume noussut 39,2 °C:een, suun avausrajoitus pahentunut (nyt vain 10 mm), oikean puolen turvotus lisääntynyt ja nieleminen vaikeutunut. Nämä löydökset viittaavat syvemmälle levinneeseen infektioprosessiin, mahdollisesti alkavaan flegmonaan tai abskessiin.
167
+
168
+ ### Arvio
169
+ Kyseessä on vakava infektio, jossa on riski hengitysteiden kompromissille ja septiselle kehitykselle. Suun avauksen rajoittuminen alle 20 mm sekä nielemisvaikeus ovat merkkejä siitä, että infektio saattaa olla levinnyt syvempiin kaulan tiloihin. Märän dreenauksen onnistuminen on kriittistä, mutta trismuksen vuoksi tämä voi olla haastavaa.
170
+
171
+ ### Suositus
172
+ Potilas tulee lähettää välittömästi päivystykseen sairaalatasolle suonensisäisen antibioottihoidon ja mahdollisen kirurgisen dreneerauksen arvioon. Hengitysteiden turvaaminen on ensisijainen prioriteetti – tarvittaessa anestesialääkärin arvio. Mikrobilääkehoitoa jatketaan sairaalassa suositusten mukaisesti. Seuranta tehostetussa valvonnassa on välttämätöntä komplikaatioiden varalta.
173
+
174
+ ## Training Details
175
+ - **Method:** LoRA (r=64, alpha=128, dropout=0.05)
176
+ - **Learning Rate:** 0.0002
177
+ - **Epochs:** 3
178
+ - **Effective Batch Size:** 64
179
+ - **Precision:** Mixed precision - mostly FP32 with some BF16 tensors, auto-casts to base model's FP16 on load
180
+ - **Embedding Training:** Enabled for domain specialization
181
+ - **Target Modules:** q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
182
+ - **Data:** 30,908 conversations covering a broad range of different scenarios
183
+ - **Context:** 2048 tokens (inherited from Ahma-3B-Instruct)
184
+ - **Hardware:** H100 GPU with Unsloth optimizations
185
+ - **Training Performance:** Loss curves showed steady convergence with final values well below 1.0, indicating successful learning without overfitting
186
+
187
+ ## Limitations and Important Notes
188
+
189
+ **Critical Medical Disclaimers:**
190
+ - Research purposes only - never use for actual medical decisions
191
+ - Model may hallucinate medical facts, treatments, or drug interactions
192
+ - Outputs have not been clinically validated and may contain medical misinformation
193
+ - All outputs require independent professional verification before any clinical application
194
+ - No liability is assumed for any consequences of model use
195
+
196
+ **Technical Limitations:**
197
+ - Primarily trained for dentist-to-dentist consultations - concept explanations included but not the main focus
198
+ - May particularly hallucinate on out-of-scope topics (further training could address this)
199
+ - Further fine-tuning may be needed for specific use cases
200
+ - System prompt dependency - using the training prompt is highly recommended
201
+ - 2048 token context limit affects both training and inference
202
+ - No formal evaluation - quality assessment has been primarily subjective
203
+
204
+ **Bias and Fairness:**
205
+ - Training data consists of synthetic conversations which may contain inherent biases
206
+ - Model responses may reflect biases present in the GPT-4.1 generated training data
207
+ - Geographic and cultural biases toward Finnish dental practices and protocols
208
+ - Potential underrepresentation of certain patient demographics or clinical scenarios
209
+
210
+ ## Related Models
211
+ - Merged version: [ducklingcodehouse/Finnish-DentalQA-merged](https://huggingface.co/ducklingcodehouse/Finnish-DentalQA-merged)
212
+ - Base model: [Finnish-NLP/Ahma-3B-Instruct](https://huggingface.co/Finnish-NLP/Ahma-3B-Instruct)
213
+
214
+ ## Citation
215
+
216
+ If you use this model, please cite both this work and the base Ahma model:
217
+
218
+ ```bibtex
219
+ @misc{finnish-dentalqa-lora,
220
+ author = {Saxén, Heikki},
221
+ title = {Finnish-DentalQA-lora: LoRA Adapter for Finnish Dental Medicine},
222
+ year = {2025},
223
+ month = {August},
224
+ publisher = {Hugging Face},
225
+ url = {https://huggingface.co/ducklingcodehouse/Finnish-DentalQA-lora}
226
+ }
227
+
228
+ @misc{ahma-3b-instruct,
229
+ author = {Tanskanen, Aapo and Toivanen, Rasmus},
230
+ title = {Ahma-3B-Instruct},
231
+ year = {2024},
232
+ publisher = {Hugging Face},
233
+ url = {https://huggingface.co/Finnish-NLP/Ahma-3B-Instruct}
234
+ }
235
+ ```
adapter_config.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "Finnish-NLP/Ahma-3B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 128,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": [
18
+ "lm_head",
19
+ "embed_tokens"
20
+ ],
21
+ "peft_type": "LORA",
22
+ "r": 64,
23
+ "rank_pattern": {},
24
+ "revision": null,
25
+ "target_modules": [
26
+ "gate_proj",
27
+ "k_proj",
28
+ "o_proj",
29
+ "up_proj",
30
+ "down_proj",
31
+ "q_proj",
32
+ "v_proj"
33
+ ],
34
+ "task_type": "CAUSAL_LM",
35
+ "use_dora": false,
36
+ "use_rslora": false
37
+ }
adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1116ace70a2d9924b9fc774ebc034bd12ecd09878e17c6cc339784153d164e4b
3
+ size 1229353552
special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<PAD>",
18
+ "lstrip": false,
19
+ "normalized": true,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": true,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<unk>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<s>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "2": {
23
+ "content": "</s>",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": false,
27
+ "single_word": false,
28
+ "special": true
29
+ },
30
+ "3": {
31
+ "content": "[INST]",
32
+ "lstrip": false,
33
+ "normalized": false,
34
+ "rstrip": false,
35
+ "single_word": false,
36
+ "special": true
37
+ },
38
+ "4": {
39
+ "content": "[/INST]",
40
+ "lstrip": false,
41
+ "normalized": false,
42
+ "rstrip": false,
43
+ "single_word": false,
44
+ "special": true
45
+ },
46
+ "5": {
47
+ "content": "<<SYS>>",
48
+ "lstrip": false,
49
+ "normalized": false,
50
+ "rstrip": false,
51
+ "single_word": false,
52
+ "special": true
53
+ },
54
+ "6": {
55
+ "content": "<</SYS>>",
56
+ "lstrip": false,
57
+ "normalized": false,
58
+ "rstrip": false,
59
+ "single_word": false,
60
+ "special": true
61
+ },
62
+ "64256": {
63
+ "content": "<PAD>",
64
+ "lstrip": false,
65
+ "normalized": true,
66
+ "rstrip": false,
67
+ "single_word": false,
68
+ "special": true
69
+ }
70
+ },
71
+ "bos_token": "<s>",
72
+ "chat_template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = 'Olet kokenut suomalainen hammaslääkäri. Vastaat ammattimaisesti kollegojesi kysymyksiin käyttäen oikeaa hammaslääketieteellistä terminologiaa ja viittaat Käypä hoito -suosituksiin kun relevanttia.' %}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 and system_message != false %}{% set content = '<<SYS>>\\n' + system_message + '\\n<</SYS>>\\n\\n' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ bos_token + ' [INST] ' + content.strip() + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' ' + content.strip() + eos_token }}{% endif %}{% endfor %}",
73
+ "clean_up_tokenization_spaces": true,
74
+ "eos_token": "</s>",
75
+ "legacy": false,
76
+ "model_max_length": 2048,
77
+ "pad_token": "<PAD>",
78
+ "padding_side": "left",
79
+ "sp_model_kwargs": {},
80
+ "spaces_between_special_tokens": false,
81
+ "tokenizer_class": "PreTrainedTokenizerFast",
82
+ "unk_token": "<unk>",
83
+ "use_default_system_prompt": false
84
+ }