mirror of
https://github.com/carlrobertoh/ProxyAI.git
synced 2026-05-19 16:28:46 +00:00
feat: Support Phi-3 Mini model (#516)
This commit is contained in:
parent
1415f387ff
commit
6d6e0a3ccb
5 changed files with 106 additions and 3 deletions
|
|
@ -52,7 +52,14 @@ public enum HuggingFaceModel {
|
|||
LLAMA_3_8B_Q8_0(8, 8, "Meta-Llama-3-8B-Instruct-Q8_0.gguf", "lmstudio-community", 8.54),
|
||||
LLAMA_3_70B_IQ1(70, 1, "Meta-Llama-3-70B-Instruct-IQ1_M.gguf", "lmstudio-community", 16.8),
|
||||
LLAMA_3_70B_IQ2_XS(70, 2, "Meta-Llama-3-70B-Instruct-IQ2_XS.gguf", "lmstudio-community", 21.1),
|
||||
LLAMA_3_70B_Q4_K_M(70, 4, "Meta-Llama-3-70B-Instruct-Q4_K_M.gguf", "lmstudio-community", 42.5);
|
||||
LLAMA_3_70B_Q4_K_M(70, 4, "Meta-Llama-3-70B-Instruct-Q4_K_M.gguf", "lmstudio-community", 42.5),
|
||||
|
||||
PHI_3_3_8B_4K_IQ4_NL(4, 4, "Phi-3-mini-4k-instruct-IQ4_NL.gguf", "lmstudio-community", 2.18),
|
||||
PHI_3_3_8B_4K_Q5_K_M(4, 5, "Phi-3-mini-4k-instruct-Q5_K_M.gguf", "lmstudio-community", 2.64),
|
||||
PHI_3_3_8B_4K_Q5_K_S(4, 5, "Phi-3-mini-4k-instruct-Q5_K_S.gguf", "lmstudio-community", 2.82),
|
||||
PHI_3_3_8B_4K_Q6_K(4, 6, "Phi-3-mini-4k-instruct-Q6_K.gguf", "lmstudio-community", 3.14),
|
||||
PHI_3_3_8B_4K_Q8_0(4, 8, "Phi-3-mini-4k-instruct-Q8_0.gguf", "lmstudio-community", 4.06),
|
||||
PHI_3_3_8B_4K_FP16(4, 16, "Phi-3-mini-4k-instruct-fp16.gguf", "lmstudio-community", 7.64);
|
||||
|
||||
private final int parameterSize;
|
||||
private final int quantization;
|
||||
|
|
|
|||
|
|
@ -99,7 +99,21 @@ public enum LlamaModel {
|
|||
HuggingFaceModel.LLAMA_3_8B_Q8_0,
|
||||
HuggingFaceModel.LLAMA_3_70B_IQ1,
|
||||
HuggingFaceModel.LLAMA_3_70B_IQ2_XS,
|
||||
HuggingFaceModel.LLAMA_3_70B_Q4_K_M));
|
||||
HuggingFaceModel.LLAMA_3_70B_Q4_K_M)),
|
||||
PHI_3(
|
||||
"Phi-3 Mini",
|
||||
"Phi-3 Mini is a 3.8B parameters, lightweight, state-of-the-art open model. "
|
||||
+ "When assessed against benchmarks testing common sense, language understanding, math, "
|
||||
+ "code, long context and logical reasoning, Phi-3 Mini-4K-Instruct showcased a robust "
|
||||
+ "and state-of-the-art performance among models with less than 13 billion parameters.",
|
||||
PromptTemplate.PHI_3,
|
||||
List.of(
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_IQ4_NL,
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_Q5_K_M,
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_Q5_K_S,
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_Q6_K,
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_Q8_0,
|
||||
HuggingFaceModel.PHI_3_3_8B_4K_FP16));
|
||||
|
||||
private final String label;
|
||||
private final String description;
|
||||
|
|
|
|||
|
|
@ -125,6 +125,25 @@ public enum PromptTemplate {
|
|||
.toString();
|
||||
}
|
||||
},
|
||||
PHI_3("Phi-3 Mini", List.of("<|end|>")) {
|
||||
@Override
|
||||
public String buildPrompt(String systemPrompt, String userPrompt, List<Message> history) {
|
||||
StringBuilder prompt = new StringBuilder();
|
||||
|
||||
for (Message message : history) {
|
||||
prompt.append("<|user|>\n")
|
||||
.append(message.getPrompt())
|
||||
.append("<|end|>\n<|assistant|>\n")
|
||||
.append(message.getResponse())
|
||||
.append("<|end|>\n");
|
||||
}
|
||||
|
||||
return prompt.append("<|user|>\n")
|
||||
.append(userPrompt)
|
||||
.append("<|end|>\n<|assistant|>")
|
||||
.toString();
|
||||
}
|
||||
},
|
||||
ALPACA("Alpaca/Vicuna") {
|
||||
@Override
|
||||
public String buildPrompt(String systemPrompt, String userPrompt, List<Message> history) {
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue