Views
No views yet
mistralai/Mistral-Small-3.1-24B-Base-2503, continued-pretrained on ≈4 B tokens of open-source military content and TIES-merged with the -Instruct variant to preserve conversational ability.
mistralai/Mistral-Small-3.1-24B-Base-2503, one epoch over the ≈4 B-token corpus, on an 8×H200 GPU cluster.mistralai/Mistral-Small-3.1-24B-Instruct-2503, to retain instruction-following and general-purpose conversational ability while keeping the domain knowledge acquired.short_answer / cloze / enum, F1 ≥ 2/3 for set, strict ordered match for permutation.| Model | Global | FR (n=314) | EN (n=247) |
|---|---|---|---|
| Gemini 3 Flash (closed, top reference) | 72.4% | 69.7% | 75.7% |
| GPT-5.4 mini (closed) | 66.7% | 61.1% | 73.7% |
| Mistral-Large 3 (≈675B) | 62.8% | 60.8% | 65.1% |
| Mistral-Small 4 (≈119B) | 58.1% | 57.3% | 59.1% |
| Mistral-Military-24B (this model) | 57.6% | 55.1% | 60.7% |
| Gemma 4 31B | 54.4% | 52.5% | 56.7% |
| Mistral-Small 3.1 (base, 24B) | 50.9% | 49.0% | 53.3% |
| Claude Haiku 4.5 | 48.1% | 48.1% | 48.2% |
| Qwen 3.5 27B | 47.8% | 44.3% | 52.2% |
Q: « Le micro-drone de contact quadrirotor utilisé en reconnaissance rapprochée par les sections d'infanterie françaises est le ____ »Gold:NX70
- Base Mistral-Small-3.1 →
Patroller(incorrect: Patroller is a MALE long-endurance drone, not a micro-drone)- Mistral-Military-24B →
NX70
Q: « Le service public chargé de l'action sociale au profit des militaires et de leurs familles (logements, crèches, vacances) est l'____ »Gold:IGESA
- Base →
Service des ressources humaines de la ministère des armées(hallucinated)- Mistral-Military-24B →
Institution de gestion sociale des armées (IGeSA)
Q: "The new French nuclear-powered attack submarine of the Barracuda programme, lead boat commissioned in 2022, is named the ____."Gold:Suffren
- Base →
SNLE 3G Le Terrible(incorrect: a ballistic-missile submarine class, not an attack submarine)- Mistral-Military-24B →
Suffren
1from transformers import AutoProcessor, AutoModelForImageTextToText
2import torch
3
4model_id = "racineai/Mistral-Military-24B"
5processor = AutoProcessor.from_pretrained(model_id)
6model = AutoModelForImageTextToText.from_pretrained(
7 model_id, dtype=torch.bfloat16, device_map="auto"
8)
9
10messages = [
11 {"role": "system", "content": [{"type": "text",
12 "text": "You are a competent assistant on defence and security topics. Answer in English."}]},
13 {"role": "user", "content": [{"type": "text",
14 "text": "Which two existing helicopter platforms is the Improved Turbine Engine Program designed to re-engine, before integration on FARA?"}]},
15]
16inputs = processor.apply_chat_template(messages, add_generation_prompt=True,
17 tokenize=True, return_dict=True,
18 return_tensors="pt").to(model.device)
19with torch.no_grad():
20 out = model.generate(**inputs, max_new_tokens=400, temperature=0.15, do_sample=True)
21print(processor.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))mistralai/Mistral-Small-3.1-24B-Base-2503, une époque sur le corpus de ≈4 milliards de tokens, sur cluster 8×H200.mistralai/Mistral-Small-3.1-24B-Instruct-2503, pour préserver les capacités de suivi d'instructions et de conversation généraliste tout en conservant les connaissances de domaine.short_answer / cloze / enum, F1 ≥ 2/3 sur set, correspondance ordonnée stricte sur permutation.| Modèle | Global | FR (n=314) | EN (n=247) |
|---|---|---|---|
| Gemini 3 Flash (closed, référence haute) | 72,4 % | 69,7 % | 75,7 % |
| GPT-5.4 mini (closed) | 66,7 % | 61,1 % | 73,7 % |
| Mistral-Large 3 (≈675 B) | 62,8 % | 60,8 % | 65,1 % |
| Mistral-Small 4 (≈119 B) | 58,1 % | 57,3 % | 59,1 % |
| Mistral-Military-24B (ce modèle) | 57,6 % | 55,1 % | 60,7 % |
| Gemma 4 31B | 54,4 % | 52,5 % | 56,7 % |
| Mistral-Small 3.1 (base, 24 B) | 50,9 % | 49,0 % | 53,3 % |
| Claude Haiku 4.5 | 48,1 % | 48,1 % | 48,2 % |
| Qwen 3.5 27B | 47,8 % | 44,3 % | 52,2 % |
Q : « Le micro-drone de contact quadrirotor utilisé en reconnaissance rapprochée par les sections d'infanterie françaises est le ____ »Réponse :NX70
- Base Mistral-Small-3.1 →
Patroller(incorrect : le Patroller est un drone MALE d'endurance, pas un micro-drone)- Mistral-Military-24B →
NX70
Q : « Le service public chargé de l'action sociale au profit des militaires et de leurs familles (logements, crèches, vacances) est l'____ »Réponse :IGESA
- Base →
Service des ressources humaines de la ministère des armées(hallucination)- Mistral-Military-24B →
Institution de gestion sociale des armées (IGeSA)
Q: "The new French nuclear-powered attack submarine of the Barracuda programme, lead boat commissioned in 2022, is named the ____."Réponse :Suffren
- Base →
SNLE 3G Le Terrible(incorrect : classe de SNLE, pas de SNA)- Mistral-Military-24B →
Suffren
1from transformers import AutoProcessor, AutoModelForImageTextToText
2import torch
3
4model_id = "racineai/Mistral-Military-24B"
5processor = AutoProcessor.from_pretrained(model_id)
6model = AutoModelForImageTextToText.from_pretrained(
7 model_id, dtype=torch.bfloat16, device_map="auto"
8)
9
10messages = [
11 {"role": "system", "content": [{"type": "text",
12 "text": "Tu es un assistant compétent en questions de défense et sécurité. Réponds en français."}]},
13 {"role": "user", "content": [{"type": "text",
14 "text": "Quelles sont les six phases de la MEDOT, dans l'ordre ?"}]},
15]
16inputs = processor.apply_chat_template(messages, add_generation_prompt=True,
17 tokenize=True, return_dict=True,
18 return_tensors="pt").to(model.device)
19with torch.no_grad():
20 out = model.generate(**inputs, max_new_tokens=400, temperature=0.15, do_sample=True)
21print(processor.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))1@misc{Mistral-Military-24B,
2 title = {Mistral-Military-24B: A French-Specialised Language Model for Defense and Security},
3 author = {{Racine.ai}},
4 year = {2026},
5 howpublished = {\url{https://huggingface.co/racineai/Mistral-Military-24B}},
6 note = {Continued pre-training of Mistral-Small-3.1-24B-Base via LoRA, then TIES-merged with Mistral-Small-3.1-24B-Instruct.}
7}Mistral-Small-3.1-24B-Base-2503 (LoRA training target) and Mistral-Small-3.1-24B-Instruct-2503 (TIES merge partner).