huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 18:48:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add Phi-3 Medium (#2205)

Browse Source
This commit is contained in:
Jani Monoses
2024-05-23 14:33:17 +03:00
committed by GitHub
parent 72e7ca529a
commit 77ea479a18
1 changed files with 13 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
candle-examples/examples/phi/main.rs
View File

@ -141,6 +141,8 @@ enum WhichModel {
V2, V2,
#[value(name = "3")] #[value(name = "3")]
V3, V3,
#[value(name = "3-medium")]
V3Medium,
#[value(name = "2-old")] #[value(name = "2-old")]
V2Old, V2Old,
PuffinPhiV2, PuffinPhiV2,
@ -254,6 +256,7 @@ fn main() -> Result<()> {
WhichModel::V1_5 => "microsoft/phi-1_5".to_string(), WhichModel::V1_5 => "microsoft/phi-1_5".to_string(),
WhichModel::V2 | WhichModel::V2Old => "microsoft/phi-2".to_string(), WhichModel::V2 | WhichModel::V2Old => "microsoft/phi-2".to_string(),
WhichModel::V3 => "microsoft/Phi-3-mini-4k-instruct".to_string(), WhichModel::V3 => "microsoft/Phi-3-mini-4k-instruct".to_string(),
WhichModel::V3Medium => "microsoft/Phi-3-medium-4k-instruct".to_string(),
WhichModel::PuffinPhiV2 | WhichModel::PhiHermes => { WhichModel::PuffinPhiV2 | WhichModel::PhiHermes => {
"lmz/candle-quantized-phi".to_string() "lmz/candle-quantized-phi".to_string()
} }
@ -273,6 +276,7 @@ fn main() -> Result<()> {
WhichModel::V2Old => "834565c23f9b28b96ccbeabe614dd906b6db551a".to_string(), WhichModel::V2Old => "834565c23f9b28b96ccbeabe614dd906b6db551a".to_string(),
WhichModel::V2 WhichModel::V2
| WhichModel::V3 | WhichModel::V3
| WhichModel::V3Medium
| WhichModel::PuffinPhiV2 | WhichModel::PuffinPhiV2
| WhichModel::PhiHermes => "main".to_string(), | WhichModel::PhiHermes => "main".to_string(),
} }
@ -287,7 +291,8 @@ fn main() -> Result<()> {
| WhichModel::V1_5 | WhichModel::V1_5
| WhichModel::V2 | WhichModel::V2
| WhichModel::V2Old | WhichModel::V2Old
| WhichModel::V3 => repo.get("tokenizer.json")?, | WhichModel::V3
| WhichModel::V3Medium => repo.get("tokenizer.json")?,
WhichModel::PuffinPhiV2 | WhichModel::PhiHermes => { WhichModel::PuffinPhiV2 | WhichModel::PhiHermes => {
repo.get("tokenizer-puffin-phi-v2.json")? repo.get("tokenizer-puffin-phi-v2.json")?
} }
@ -303,14 +308,14 @@ fn main() -> Result<()> {
WhichModel::V2 | WhichModel::V2Old => vec![repo.get("model-v2-q4k.gguf")?], WhichModel::V2 | WhichModel::V2Old => vec![repo.get("model-v2-q4k.gguf")?],
WhichModel::PuffinPhiV2 => vec![repo.get("model-puffin-phi-v2-q4k.gguf")?], WhichModel::PuffinPhiV2 => vec![repo.get("model-puffin-phi-v2-q4k.gguf")?],
WhichModel::PhiHermes => vec![repo.get("model-phi-hermes-1_3B-q4k.gguf")?], WhichModel::PhiHermes => vec![repo.get("model-phi-hermes-1_3B-q4k.gguf")?],
WhichModel::V3 => anyhow::bail!( WhichModel::V3 | WhichModel::V3Medium => anyhow::bail!(
"use the quantized or quantized-phi examples for quantized phi-v3" "use the quantized or quantized-phi examples for quantized phi-v3"
), ),
} }
} else { } else {
match args.model { match args.model {
WhichModel::V1 | WhichModel::V1_5 => vec![repo.get("model.safetensors")?], WhichModel::V1 | WhichModel::V1_5 => vec![repo.get("model.safetensors")?],
WhichModel::V2 | WhichModel::V2Old | WhichModel::V3 => { WhichModel::V2 | WhichModel::V2Old | WhichModel::V3 | WhichModel::V3Medium => {
candle_examples::hub_load_safetensors( candle_examples::hub_load_safetensors(
&repo, &repo,
"model.safetensors.index.json", "model.safetensors.index.json",
@ -332,7 +337,7 @@ fn main() -> Result<()> {
WhichModel::V2 | WhichModel::V2Old => Config::v2(), WhichModel::V2 | WhichModel::V2Old => Config::v2(),
WhichModel::PuffinPhiV2 => Config::puffin_phi_v2(), WhichModel::PuffinPhiV2 => Config::puffin_phi_v2(),
WhichModel::PhiHermes => Config::phi_hermes_1_3b(), WhichModel::PhiHermes => Config::phi_hermes_1_3b(),
WhichModel::V3 => { WhichModel::V3 | WhichModel::V3Medium => {
panic!("use the quantized or quantized-phi examples for quantized phi-v3") panic!("use the quantized or quantized-phi examples for quantized phi-v3")
} }
}; };
@ -352,7 +357,9 @@ fn main() -> Result<()> {
let dtype = match args.dtype { let dtype = match args.dtype {
Some(dtype) => std::str::FromStr::from_str(&dtype)?, Some(dtype) => std::str::FromStr::from_str(&dtype)?,
None => { None => {
if args.model == WhichModel::V3 && device.is_cuda() { if (args.model == WhichModel::V3 || args.model == WhichModel::V3Medium)
&& device.is_cuda()
{
DType::BF16 DType::BF16
} else { } else {
DType::F32 DType::F32
@ -368,7 +375,7 @@ fn main() -> Result<()> {
let phi = Phi::new(&config, vb)?; let phi = Phi::new(&config, vb)?;
Model::Phi(phi) Model::Phi(phi)
} }
WhichModel::V3 => { WhichModel::V3 | WhichModel::V3Medium => {
let config_filename = repo.get("config.json")?; let config_filename = repo.get("config.json")?;
let config = std::fs::read_to_string(config_filename)?; let config = std::fs::read_to_string(config_filename)?;
let config: Phi3Config = serde_json::from_str(&config)?; let config: Phi3Config = serde_json::from_str(&config)?;