vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add new hf protocol for ollama (#11449)

Browse source 
https://huggingface.co/docs/hub/en/ollama

Signed-off-by: Eric Curtin <ecurtin@redhat.com>
This commit is contained in:
Eric Curtin 2025-01-27 19:36:10 +01:00 • committed by GitHub
parent d6d24cd9ed
commit a4417ddda9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 80 additions and 41 deletions
Download patch file Download diff file Expand all files Collapse all files

121
examples/run/run.cpp
View file

@ -319,6 +319,10 @@ class HttpClient {
public: public:
int init(const std::string & url, const std::vector<std::string> & headers, const std::string & output_file, int init(const std::string & url, const std::vector<std::string> & headers, const std::string & output_file,
const bool progress, std::string * response_str = nullptr) { const bool progress, std::string * response_str = nullptr) {
if (std::filesystem::exists(output_file)) {
return 0;
}
std::string output_file_partial; std::string output_file_partial;
curl = curl_easy_init(); curl = curl_easy_init();
if (!curl) { if (!curl) {
@ -558,13 +562,14 @@ class LlamaData {
} }
sampler = initialize_sampler(opt); sampler = initialize_sampler(opt);
return 0; return 0;
} }
private: private:
#ifdef LLAMA_USE_CURL #ifdef LLAMA_USE_CURL
int download(const std::string & url, const std::vector<std::string> & headers, const std::string & output_file, int download(const std::string & url, const std::string & output_file, const bool progress,
const bool progress, std::string * response_str = nullptr) { const std::vector<std::string> & headers = {}, std::string * response_str = nullptr) {
HttpClient http; HttpClient http;
if (http.init(url, headers, output_file, progress, response_str)) { if (http.init(url, headers, output_file, progress, response_str)) {
return 1; return 1;
@ -573,48 +578,85 @@ class LlamaData {
return 0; return 0;
} }
#else #else
int download(const std::string &, const std::vector<std::string> &, const std::string &, const bool, int download(const std::string &, const std::string &, const bool, const std::vector<std::string> & = {},
std::string * = nullptr) { std::string * = nullptr) {
printe("%s: llama.cpp built without libcurl, downloading from an url not supported.\n", __func__); printe("%s: llama.cpp built without libcurl, downloading from an url not supported.\n", __func__);
return 1; return 1;
} }
#endif #endif
int huggingface_dl(const std::string & model, const std::vector<std::string> headers, const std::string & bn) { // Helper function to handle model tag extraction and URL construction
// Find the second occurrence of '/' after protocol string std::pair<std::string, std::string> extract_model_and_tag(std::string & model, const std::string & base_url) {
size_t pos = model.find('/'); std::string model_tag = "latest";
pos = model.find('/', pos + 1); const size_t colon_pos = model.find(':');
if (pos == std::string::npos) {
return 1;
}
const std::string hfr = model.substr(0, pos);
const std::string hff = model.substr(pos + 1);
const std::string url = "https://huggingface.co/" + hfr + "/resolve/main/" + hff;
return download(url, headers, bn, true);
}
int ollama_dl(std::string & model, const std::vector<std::string> headers, const std::string & bn) {
if (model.find('/') == std::string::npos) {
model = "library/" + model;
}
std::string model_tag = "latest";
size_t colon_pos = model.find(':');
if (colon_pos != std::string::npos) { if (colon_pos != std::string::npos) {
model_tag = model.substr(colon_pos + 1); model_tag = model.substr(colon_pos + 1);
model = model.substr(0, colon_pos); model = model.substr(0, colon_pos);
} }
std::string manifest_url = "https://registry.ollama.ai/v2/" + model + "/manifests/" + model_tag; std::string url = base_url + model + "/manifests/" + model_tag;
return { model, url };
}
// Helper function to download and parse the manifest
int download_and_parse_manifest(const std::string & url, const std::vector<std::string> & headers,
nlohmann::json & manifest) {
std::string manifest_str; std::string manifest_str;
const int ret = download(manifest_url, headers, "", false, &manifest_str); int ret = download(url, "", false, headers, &manifest_str);
if (ret) { if (ret) {
return ret; return ret;
} }
nlohmann::json manifest = nlohmann::json::parse(manifest_str); manifest = nlohmann::json::parse(manifest_str);
std::string layer;
return 0;
}
int huggingface_dl(std::string & model, const std::string & bn) {
// Find the second occurrence of '/' after protocol string
size_t pos = model.find('/');
pos = model.find('/', pos + 1);
std::string hfr, hff;
std::vector<std::string> headers = { "User-Agent: llama-cpp", "Accept: application/json" };
std::string url;
if (pos == std::string::npos) {
auto [model_name, manifest_url] = extract_model_and_tag(model, "https://huggingface.co/v2/");
hfr = model_name;
nlohmann::json manifest;
int ret = download_and_parse_manifest(manifest_url, headers, manifest);
if (ret) {
return ret;
}
hff = manifest["ggufFile"]["rfilename"];
} else {
hfr = model.substr(0, pos);
hff = model.substr(pos + 1);
}
url = "https://huggingface.co/" + hfr + "/resolve/main/" + hff;
return download(url, bn, true, headers);
}
int ollama_dl(std::string & model, const std::string & bn) {
const std::vector<std::string> headers = { "Accept: application/vnd.docker.distribution.manifest.v2+json" };
if (model.find('/') == std::string::npos) {
model = "library/" + model;
}
auto [model_name, manifest_url] = extract_model_and_tag(model, "https://registry.ollama.ai/v2/");
nlohmann::json manifest;
int ret = download_and_parse_manifest(manifest_url, {}, manifest);
if (ret) {
return ret;
}
std::string layer;
for (const auto & l : manifest["layers"]) { for (const auto & l : manifest["layers"]) {
if (l["mediaType"] == "application/vnd.ollama.image.model") { if (l["mediaType"] == "application/vnd.ollama.image.model") {
layer = l["digest"]; layer = l["digest"];
@ -622,8 +664,9 @@ class LlamaData {
} }
} }
std::string blob_url = "https://registry.ollama.ai/v2/" + model + "/blobs/" + layer; std::string blob_url = "https://registry.ollama.ai/v2/" + model_name + "/blobs/" + layer;
return download(blob_url, headers, bn, true);
return download(blob_url, bn, true, headers);
} }
std::string basename(const std::string & path) { std::string basename(const std::string & path) {
@ -653,22 +696,18 @@ class LlamaData {
return ret; return ret;
} }
const std::string bn = basename(model_); const std::string bn = basename(model_);
const std::vector<std::string> headers = { "--header",
"Accept: application/vnd.docker.distribution.manifest.v2+json" };
if (string_starts_with(model_, "hf://") || string_starts_with(model_, "huggingface://")) { if (string_starts_with(model_, "hf://") || string_starts_with(model_, "huggingface://")) {
rm_until_substring(model_, "://"); rm_until_substring(model_, "://");
ret = huggingface_dl(model_, headers, bn); ret = huggingface_dl(model_, bn);
} else if (string_starts_with(model_, "hf.co/")) { } else if (string_starts_with(model_, "hf.co/")) {
rm_until_substring(model_, "hf.co/"); rm_until_substring(model_, "hf.co/");
ret = huggingface_dl(model_, headers, bn); ret = huggingface_dl(model_, bn);
} else if (string_starts_with(model_, "ollama://")) {
rm_until_substring(model_, "://");
ret = ollama_dl(model_, headers, bn);
} else if (string_starts_with(model_, "https://")) { } else if (string_starts_with(model_, "https://")) {
ret = download(model_, headers, bn, true); ret = download(model_, bn, true);
} else { } else { // ollama:// or nothing
ret = ollama_dl(model_, headers, bn); rm_until_substring(model_, "://");
ret = ollama_dl(model_, bn);
} }
model_ = bn; model_ = bn;