Instructions to use StandardThinking/StandardOne-3B-SH-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use StandardThinking/StandardOne-3B-SH-GGUF with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M # Run inference directly in the terminal: llama cli -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M # Run inference directly in the terminal: llama cli -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama cli -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Use Docker
docker model run hf.co/StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
- Llama new
- LM Studio
- Jan
- Ollama
How to use StandardThinking/StandardOne-3B-SH-GGUF with Ollama:
ollama run hf.co/StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
- Unsloth Desktop
- Pi
How to use StandardThinking/StandardOne-3B-SH-GGUF with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @earendil-works/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use StandardThinking/StandardOne-3B-SH-GGUF with Docker Model Runner:
docker model run hf.co/StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
- Lemonade
How to use StandardThinking/StandardOne-3B-SH-GGUF with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Run and chat with the model
lemonade run user.StandardOne-3B-SH-GGUF-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use StandardThinking/StandardOne-3B-SH-GGUF with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:9931/v1 hermes config set model.default StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use StandardThinking/StandardOne-3B-SH-GGUF with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:9931/v1 \ --custom-model-id "StandardThinking/StandardOne-3B-SH-GGUF:Q4_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
File size: 6,758 Bytes
eb6c5ec | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 | // sh-states: dump the hidden states a Standard One schema head reads (final norm output + one decoder layer output)
// from a llama.cpp GGUF model, for already-tokenized inputs.
//
// sh-states -m model.gguf --tap 25 [-t threads] [-c n_ctx] [-b chunk]
//
// Protocol (line based, stdin -> stdout): each request line is
// <ids.bin> <out_prefix>
// where ids.bin holds little-endian int32 token ids (BOS included, exactly as the renderer produced them). For every
// request the program writes <out_prefix>.final.f32 and <out_prefix>.tap.f32 (row-major float32 [n_tokens, n_embd])
// and answers "ok <n_tokens> <n_embd>" (or "err <message>"). The KV cache is cleared between requests.
//
// The final state is llama.cpp's per-token embedding output (embeddings=true, pooling NONE: the final RMSNorm output =
// HF last_hidden_state). The tap is taken with the scheduler eval callback on "l_out-<tap>" (= the output of decoder
// layer <tap>, as a forward hook on layers[tap] sees it). The prompt is decoded in chunks with every token marked as
// an output, so both cover all tokens.
#include "llama.h"
#include "ggml.h"
#include "ggml-backend.h"
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <fstream>
#include <iostream>
#include <sstream>
#include <string>
#include <vector>
struct capture {
std::string tap_name;
std::vector<float> final_rows, tap_rows;
int64_t n_embd = 0;
};
static bool eval_cb(struct ggml_tensor * t, bool ask, void * user_data) {
auto * cap = (capture *) user_data;
const bool want = cap->tap_name == t->name;
if (ask) {
return want;
}
if (!want) {
return true;
}
if (t->type != GGML_TYPE_F32 || ggml_n_dims(t) > 2) {
fprintf(stderr, "unexpected tensor %s type %d dims %d\n", t->name, (int) t->type, ggml_n_dims(t));
return false;
}
const int64_t n_embd = t->ne[0], n_tok = t->ne[1];
cap->n_embd = n_embd;
std::vector<float> buf(n_embd * n_tok);
if (t->nb[1] == (size_t) n_embd * sizeof(float)) {
ggml_backend_tensor_get(t, buf.data(), 0, buf.size() * sizeof(float));
} else { // non-contiguous rows
for (int64_t r = 0; r < n_tok; r++) {
ggml_backend_tensor_get(t, buf.data() + r * n_embd, r * t->nb[1], n_embd * sizeof(float));
}
}
cap->tap_rows.insert(cap->tap_rows.end(), buf.begin(), buf.end());
return true;
}
static bool write_f32(const std::string & path, const std::vector<float> & v) {
std::ofstream f(path, std::ios::binary);
f.write((const char *) v.data(), v.size() * sizeof(float));
return (bool) f;
}
int main(int argc, char ** argv) {
std::string model_path;
int tap = 25, threads = 8, n_ctx = 34000, chunk = 512;
for (int i = 1; i < argc; i++) {
std::string a = argv[i];
auto next = [&]() { return std::string(i + 1 < argc ? argv[++i] : ""); };
if (a == "-m") model_path = next();
else if (a == "--tap") tap = std::stoi(next());
else if (a == "-t") threads = std::stoi(next());
else if (a == "-c") n_ctx = std::stoi(next());
else if (a == "-b") chunk = std::stoi(next());
else { fprintf(stderr, "unknown argument %s\n", a.c_str()); return 2; }
}
if (model_path.empty()) { fprintf(stderr, "usage: sh-states -m model.gguf --tap 25\n"); return 2; }
llama_backend_init();
llama_model_params mp = llama_model_default_params();
llama_model * model = llama_model_load_from_file(model_path.c_str(), mp);
if (!model) { fprintf(stderr, "failed to load %s\n", model_path.c_str()); return 1; }
if (tap < 0 || tap >= llama_model_n_layer(model)) { fprintf(stderr, "tap out of range\n"); return 2; }
capture cap;
cap.tap_name = "l_out-" + std::to_string(tap);
llama_context_params cp = llama_context_default_params();
cp.n_ctx = n_ctx;
cp.n_batch = chunk;
cp.n_ubatch = chunk;
cp.n_threads = threads;
cp.n_threads_batch = threads;
cp.embeddings = true;
cp.pooling_type = LLAMA_POOLING_TYPE_NONE;
cp.cb_eval = eval_cb;
cp.cb_eval_user_data = ∩
llama_context * ctx = llama_init_from_model(model, cp);
if (!ctx) { fprintf(stderr, "failed to create context\n"); return 1; }
printf("ready n_layer=%d n_embd=%d tap=%d\n", llama_model_n_layer(model), llama_model_n_embd(model), tap);
fflush(stdout);
std::string line;
while (std::getline(std::cin, line)) {
std::istringstream ss(line);
std::string ids_path, out_prefix;
if (!(ss >> ids_path >> out_prefix)) { printf("err bad request line\n"); fflush(stdout); continue; }
std::ifstream f(ids_path, std::ios::binary | std::ios::ate);
if (!f) { printf("err cannot read %s\n", ids_path.c_str()); fflush(stdout); continue; }
const size_t bytes = f.tellg();
f.seekg(0);
std::vector<llama_token> ids(bytes / sizeof(int32_t));
f.read((char *) ids.data(), bytes);
const int n = (int) ids.size();
if (n == 0 || n > n_ctx) { printf("err %d tokens (context %d)\n", n, n_ctx); fflush(stdout); continue; }
llama_memory_clear(llama_get_memory(ctx), true);
cap.final_rows.clear();
cap.tap_rows.clear();
bool ok = true;
llama_batch batch = llama_batch_init(chunk, 0, 1);
for (int s = 0; s < n && ok; s += chunk) {
const int m = std::min(chunk, n - s);
batch.n_tokens = m;
for (int j = 0; j < m; j++) {
batch.token[j] = ids[s + j];
batch.pos[j] = s + j;
batch.n_seq_id[j] = 1;
batch.seq_id[j][0] = 0;
batch.logits[j] = 1;
}
if (llama_decode(ctx, batch) != 0) { ok = false; break; }
const float * emb = llama_get_embeddings(ctx);
if (!emb) { ok = false; break; }
const int64_t h = llama_model_n_embd(model);
cap.final_rows.insert(cap.final_rows.end(), emb, emb + (size_t) m * h);
}
llama_batch_free(batch);
const size_t want = (size_t) n * cap.n_embd;
if (!ok || cap.final_rows.size() != want || cap.tap_rows.size() != want) {
printf("err decode=%d final=%zu tap=%zu want=%zu\n", ok, cap.final_rows.size(), cap.tap_rows.size(), want);
fflush(stdout);
continue;
}
if (!write_f32(out_prefix + ".final.f32", cap.final_rows) || !write_f32(out_prefix + ".tap.f32", cap.tap_rows)) {
printf("err write failed\n");
} else {
printf("ok %d %lld\n", n, (long long) cap.n_embd);
}
fflush(stdout);
}
llama_free(ctx);
llama_model_free(model);
llama_backend_free();
return 0;
}
|