Skip to content

Commit

Permalink
fix(doc/examples): set defaults to mirostat (#1820)
Browse files Browse the repository at this point in the history
The default sampler on some models don't return enough candidates which
leads to a false sense of randomness. Tracing back the code it looks
that with the temperature sampler there might not be enough
candidates to pick from, and since the seed and "randomness" take effect
while picking a good candidate this yields to the same results over and
over.

Fixes #1723 by updating the
examples and documentation to use mirostat instead.
  • Loading branch information
mudler authored Mar 11, 2024
1 parent 8e57f4d commit bc8f648
Show file tree
Hide file tree
Showing 7 changed files with 28 additions and 5 deletions.
7 changes: 4 additions & 3 deletions docs/content/docs/advanced/advanced-usage.md
Original file line number Diff line number Diff line change
Expand Up @@ -130,13 +130,14 @@ parameters:
typical_p:
tfz:
frequency_penalty:
mirostat_eta:
mirostat_tau:
mirostat:
rope_freq_base:
rope_freq_scale:
negative_prompt_scale:
mirostat_eta:
mirostat_tau:
mirostat:
# Default context size
context_size: 512
# Default number of threads
Expand Down
6 changes: 5 additions & 1 deletion embedded/models/codellama-7b-gguf.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,14 @@ name: codellama-7b-gguf
backend: transformers
parameters:
model: huggingface://TheBloke/CodeLlama-7B-GGUF/codellama-7b.Q4_K_M.gguf
temperature: 0.2
temperature: 0.5
top_k: 40
seed: -1
top_p: 0.95
mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0

context_size: 4096
f16: true
gpu_layers: 90
Expand Down
5 changes: 4 additions & 1 deletion embedded/models/dolphin-2.5-mixtral-8x7b.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,13 @@ name: dolphin-mixtral-8x7b
mmap: true
parameters:
model: huggingface://TheBloke/dolphin-2.5-mixtral-8x7b-GGUF/dolphin-2.5-mixtral-8x7b.Q2_K.gguf
temperature: 0.2
temperature: 0.5
top_k: 40
top_p: 0.95
seed: -1
mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0
template:
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "user"}}user{{end}}
Expand Down
3 changes: 3 additions & 0 deletions embedded/models/llava.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,9 @@ parameters:
top_k: 40
top_p: 0.95
seed: -1
mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0

template:
chat: |
Expand Down
4 changes: 4 additions & 0 deletions embedded/models/mistral-openorca.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,10 @@ parameters:
top_k: 40
top_p: 0.95
seed: -1
mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0

template:
chat_message: |
<|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "user"}}user{{end}}
Expand Down
4 changes: 4 additions & 0 deletions embedded/models/mixtral-instruct.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,10 @@ parameters:
top_k: 40
seed: -1
top_p: 0.95
mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0

template:
chat: &chat |
[INST] {{.Input}} [/INST]
Expand Down
4 changes: 4 additions & 0 deletions examples/configurations/phi-2.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,10 @@ parameters:
top_k: 40
top_p: 0.95
seed: -1

mirostat: 2
mirostat_eta: 1.0
mirostat_tau: 1.0
template:
chat: &template |-
Instruct: {{.Input}}
Expand Down

0 comments on commit bc8f648

Please sign in to comment.