MichelRosselli/ emma-6:q8_0

41 3 days ago

Emma-6 is an open-source Italian Large Language Model developed by Egomnia S.p.A. 

ollama run MichelRosselli/emma-6:q8_0

Details

4 days ago

838d62eca7c8 · 1.6GB

model
llama
·
1.55B
·
Q8_0
system
Sei Emma-6, l'assistente AI di Egomnia S.p.A. Regole: 1. Rispondi sempre in italiano. 2. Sii preciso
params
{ "min_p": 0, "num_ctx": 4096, "repeat_last_n": 4096, "repeat_penalty": 1.1, "st
template
{{- range $i, $m := .Messages }} {{- $last := eq (len (slice $.Messages $i)) 1 }} {{- $c := $m.Conte

Readme

Emma-6

Unofficial Ollama version of Emma-6, the Italian language model developed by Egomnia S.p.A.

ollama run MichelRosselli/emma-6

latest is the full-precision version (F32). Smaller tags: f16, bf16, q8_0, q6_k, q5_k_m, q4_k_m, q3_k_m, q2_k.

  • The chat format, the default sampling parameters (official “Bilanciata” preset: temperature 0.7, top-k 40, top-p 0.9, repeat penalty 1.1) and the 4096-token context come from Egomnia’s official service.
  • The default system prompt is not the one of the official service, which Egomnia keeps reserved: it is the alternative Egomnia published for local use. Replace it with your own system message (/set system ...), or remove it with an empty one.
  • It handles short conversations: for the most reliable results keep the history short, or send one complete request at a time.
  • Line breaks in your messages reach the model as they are, while the original tokenizer turned them into spaces. For the best results write each message on a single line.
  • Replies may start with a space: this comes from token-by-token decoding and does not change the answer.

Links:

License

Released under the OpenRAIL license, like the original model: refer to the license terms distributed with the model before using, modifying or redistributing it. Authorship of the model belongs to Egomnia S.p.A.; please credit it when you use the model in products, services, publications or technical documentation.