| --- |
| language: |
| - ml |
| - en |
| license: apache-2.0 |
| tags: |
| - malayalam |
| - indic-llm |
| - instruction-tuned |
| - text-generation |
| - kerala |
| pipeline_tag: text-generation |
| --- |
| |
| # 🪔 Guru (ഗുരു) |
|
|
| **ഗുരു** എന്നാൽ "അധ്യാപകൻ" അല്ലെങ്കിൽ "വഴികാട്ടി" എന്നാണർത്ഥം. ഈ പേരു അന്വർത്ഥമാക്കുന്ന |
| തരത്തിൽ, **Guru** മലയാളത്തിൽ സ്വാഭാവികമായും കൃത്യമായും ആശയവിനിമയം നടത്താൻ കഴിവുള്ള ഒരു |
| ഭാഷാ മോഡലാണ്. കേരള സർക്കാർ സേവനങ്ങളെക്കുറിച്ചുള്ള ചോദ്യങ്ങൾക്ക് ഉത്തരം നൽകാനും |
| പൊതുവായ മലയാളം നിർദ്ദേശങ്ങൾ പിന്തുടരാനും ഇത് പ്രത്യേകം പരിശീലിപ്പിച്ചിരിക്കുന്നു. |
|
|
| **Guru** is a Malayalam-focused instruction-tuned language model, built to understand and |
| generate natural, fluent Malayalam — with particular strength in answering questions about |
| Kerala government services and general Malayalam instruction-following. |
|
|
| ## Model Details |
|
|
| - **Model name:** Guru (`NeoSapian/GURU-v1`) |
| - **Language focus:** Malayalam (ml), with English support |
| - **Domain:** Kerala government services, general Malayalam instruction-following |
| - **Format:** Full-precision merged weights (safetensors) |
|
|
| ## Training Data |
|
|
| Fine-tuned on a curated instruction dataset of Malayalam prompt/response pairs covering |
| Kerala government services, formatted in an Alpaca / role-prompt style. |
|
|
| ## Intended Use |
|
|
| - Malayalam-language chat and instruction-following assistants |
| - Answering questions about Kerala government services in Malayalam |
| - General Malayalam NLP research and experimentation |
|
|
| ## How to Use |
|
|
| ```python |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
| |
| model_id = "NeoSapian/GURU-v1" |
| tokenizer = AutoTokenizer.from_pretrained(model_id) |
| model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="auto") |
| |
| messages = [{"role": "user", "content": "കേരള സർക്കാരിന്റെ ക്ഷേമ പദ്ധതികളെക്കുറിച്ച് പറയാമോ?"}] |
| inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device) |
| outputs = model.generate(inputs, max_new_tokens=512) |
| print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
| ``` |
|
|
| ## Limitations |
|
|
| - Fine-tuned on a domain-specific dataset (Kerala govt services), so performance on topics outside this domain may vary. |
| - Not evaluated for safety-critical, legal, or medical decision-making — verify outputs before relying on them for such use cases. |
|
|
| ## License |
|
|
| Released under **Apache 2.0**. |
|
|
| ## Citation |
|
|
| If you use Guru, please consider citing this repository. |
|
|
