Intrenion

LLM Parameters

Christian Ullrich
June 2025

LLM parameters for a ChatGPT-like experience

GPT4All

Parameter Phi-4-14B
Context Length 4096
Max Length 4096
Prompt Batch Size 128
Temperature 0,7
Top-P 0,9
Top-K 40
Min-P 0
Repeat Penalty Tokens 256
GPU Layers 40
Repeat Penalty 1,1

Text generation web UI

Category Setting Phi-4-14B Qwen3-32B
Curve shape temperature 0,7 0,7
Curve cutoff min_p 0 0
  top_n_sigma 1 1
  top_p 0,9 0,9
  top_k 40 40
  typical_p 1 1
  xtc_threshold 0 0
  xtc_probability 0 0
Repetition suppression dry_multiplier 1 1
  dry_allowed_length 1 1
  dry_base 1 1
  repetition_penalty 1,1 1,2
  frequency_penalty 0 0
  presence_penalty 0 0
  repetition_penalty_range 256 256
Alternative sampling mirostat_mode 0 0
  mirostat_tau 0 0
  mirostat_eta 0 0
Other options max_new_tokens 512 1024
  Maximum tokens/second 0 0
  Maximum UI updates/second 5 5
Other settings 1 dynamic temperature ❌ (deactivated) ❌ (deactivated)
  temperature_last ❌ (deactivated) ❌ (deactivated)
  auto_max_new_tokens ❌ (deactivated) ❌ (deactivated)
  Ban the eos_token ❌ (deactivated) ❌ (deactivated)
  Add the bos_token to the beginning of prompts ✅ (activated) ✅ (activated)
  Enable thinking ❌ (deactivated) ❌ (deactivated)
  Activate text streaming ✅ (activated) ✅ (activated)
Other settings 2 Truncate the prompt up to this length 4096 8192
  Seed (-1 for random) -1 -1

Homepage - Terms of service • Privacy policy • Legal notice