frob/ hy4-preview:770b-a49b

31 2 weeks ago

770b
ollama run frob/hy4-preview:770b-a49b

Details

2 weeks ago

c4c95e6e9ea6 · 465GB

hy_v4
·
770B
·
Q4_K_M
Tencent is pleased to support the open source community by making Tencent Hy4 preview available. Cop
{ "temperature": 0.9, "top_p": 1 }

Readme

Imported from https://hf.co/tencent/Hy4-preview

Requires ollama v0.34.1+.

Chat/thinking

The model supports high and no_think levels of reasoning, unfortunately ollama doesn’t accept no_think as a valid level so currently the model always uses high thinking.

$ ollama run frob/hy4-preview
>>> hello
Thinking...
We are in a new conversation. The user said "hello". I should respond 
appropriately, likely with a greeting and an offer to help. I am an AI 
assistant, so I should be helpful and harmless. Let's craft a friendly 
response.

I need to introduce myself? Not necessarily, but I can say hello back and 
ask how I can help.

Let's respond: "Hello! How can I help you today?" or something similar.

Keep it concise and friendly.
...done thinking.

Hello! How can I help you today?

>>> /set think no_think
warning: model "frob/hy4-preview" does not support thinking output
Set 'think' mode to 'no_think'.
>>> /clear
Cleared session context
>>> hello
error: 400 Bad Request: invalid think value: "no_think" (must be "high", "medium", "low", "max", true, or false)
>>> /set nothink
warning: model "frob/hy4-preview" does not support thinking output
Set 'nothink' mode.
>>> hello
Thinking...
We need respond to user. They sent "hello" multiple times. We need be 
helpful, acknowledge. Could be test or repetition. We should say hello back 
and ask how can help. No need mention repetition unless helpful. Since 
repeated hello, maybe respond with a friendly greeting and maybe note 
they've said hello several times, ask if they need assistance. Be concise.

We need ensure no analysis in final except needed. Final: "Hello! I see 
you've said hello a few times — how can I help you today?" Maybe that's 
good. But maybe avoid being snarky. Use calm tone.

We can also just say "Hello! How can I assist you today?" The repetition 
might be accidental. Maybe include light acknowledgment: "Hello! Looks like 
you said hello several times — how can I help?" That is friendly.

Let's craft final.
...done thinking.

Hello! It looks like you said hello a few times — how can I help you today?

Tools

$ ollama launch opencode \
    --model frob/hy4-preview \
    -- \
    run 'create a file called README with the contents "hello world"'

> build · frob/hy4-preview:latest

← Write README
Wrote file successfully.

Created `README` in `/home/rick/ollama/issues/hy4-preview`.
$ cat README 
hello world