matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 07:46:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

convert safetensor adapters into GGUF (#6327)

Browse Source
This commit is contained in:
Patrick Devine
2024-08-23 11:29:56 -07:00
committed by GitHub
parent 7a1e1c1caf
commit 0c819e167b
16 changed files with 697 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
convert/convert_mixtral.go
View File

@@ -9,14 +9,14 @@ import (
"github.com/ollama/ollama/llm"
)
type mixtral struct {
llama
type mixtralModel struct {
llamaModel
NumLocalExperts uint32 `json:"num_local_experts"`
NumExpertsPerToken uint32 `json:"num_experts_per_tok"`
}
func (p *mixtral) KV(t *Tokenizer) llm.KV {
kv := p.llama.KV(t)
func (p *mixtralModel) KV(t *Tokenizer) llm.KV {
kv := p.llamaModel.KV(t)
if p.NumLocalExperts > 0 {
kv["llama.expert_count"] = p.NumLocalExperts
@@ -29,7 +29,7 @@ func (p *mixtral) KV(t *Tokenizer) llm.KV {
return kv
}
func (p *mixtral) Tensors(ts []Tensor) []llm.Tensor {
func (p *mixtralModel) Tensors(ts []Tensor) []llm.Tensor {
oldnew := []string{
"model.layers", "blk",
"w1", "ffn_gate_exps",
@@ -67,12 +67,12 @@ func (p *mixtral) Tensors(ts []Tensor) []llm.Tensor {
})
}
return append(out, p.llama.Tensors(ts)...)
return append(out, p.llamaModel.Tensors(ts)...)
}
func (p *mixtral) Replacements() []string {
func (p *mixtralModel) Replacements() []string {
return append(
p.llama.Replacements(),
p.llamaModel.Replacements(),
"block_sparse_moe.gate", "ffn_gate_inp",
)
}