Text Classification
GGUF
English
rknn
rknn3
rk1828
rk1820
rockchip
npu
edge-ai
kev
decision
w4a16
Eval Results (legacy)
conversational
Instructions to use jj90/Kev-0.8B-RKNN3-RK1828 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 jj90/Kev-0.8B-RKNN3-RK1828 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 jj90/Kev-0.8B-RKNN3-RK1828 # Run inference directly in the terminal: llama cli -hf jj90/Kev-0.8B-RKNN3-RK1828
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf jj90/Kev-0.8B-RKNN3-RK1828 # Run inference directly in the terminal: llama cli -hf jj90/Kev-0.8B-RKNN3-RK1828
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 jj90/Kev-0.8B-RKNN3-RK1828 # Run inference directly in the terminal: ./build/bin/llama cli -hf jj90/Kev-0.8B-RKNN3-RK1828
Use Docker
docker model run hf.co/jj90/Kev-0.8B-RKNN3-RK1828
- Llama new
- LM Studio
- Jan
- Ollama
How to use jj90/Kev-0.8B-RKNN3-RK1828 with Ollama:
ollama run hf.co/jj90/Kev-0.8B-RKNN3-RK1828
- Unsloth Desktop
- Pi
How to use jj90/Kev-0.8B-RKNN3-RK1828 with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf jj90/Kev-0.8B-RKNN3-RK1828
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": "jj90/Kev-0.8B-RKNN3-RK1828" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use jj90/Kev-0.8B-RKNN3-RK1828 with Docker Model Runner:
docker model run hf.co/jj90/Kev-0.8B-RKNN3-RK1828
- Lemonade
How to use jj90/Kev-0.8B-RKNN3-RK1828 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull jj90/Kev-0.8B-RKNN3-RK1828
Run and chat with the model
lemonade run user.Kev-0.8B-RKNN3-RK1828-{{QUANT_TAG}}List all available models
lemonade list
- Hermes Agent
How to use jj90/Kev-0.8B-RKNN3-RK1828 with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf jj90/Kev-0.8B-RKNN3-RK1828
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 jj90/Kev-0.8B-RKNN3-RK1828
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use jj90/Kev-0.8B-RKNN3-RK1828 with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf jj90/Kev-0.8B-RKNN3-RK1828
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 "jj90/Kev-0.8B-RKNN3-RK1828" \ --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: 15,367 Bytes
a846b3d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 | // Copyright (c) 2026 by Rockchip Electronics Co., Ltd. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "rknn3_session_test_common.h"
// ---- 我们加的:output_callback 里直接拿到 logits 张量并落盘 ----
// sampling_callback 不会被调用(运行时内部采样),但 output_tensors[0] 就是 vocabulary logits
static int g_kev_dumped = 0;
static int g_kev_seq = 0; // 当前第几条序列,用来给 logits 文件编号
static char g_out_path[512] = "D:\\rk1828-console\\kevrunner\\out\\cur.bin";
static int kev_output_callback(void* userdata, rknn3_tensor* outs, uint32_t n,
LLMOutputCallbackState state) {
(void)userdata;
if (!g_kev_dumped && n > 0 && state == RKLLM_OUTPUT_CALLBACK_PREFILL_FINISHED) {
const char* p = g_out_path;
FILE* f = fopen(p, "wb");
if (f) {
size_t sz = outs[0].mem->size;
fwrite(outs[0].mem->virt_addr, 1, sz, f);
fclose(f);
printf("[kev] dumped %zu bytes, state=%d, name=%s\n", sz, (int)state,
outs[0].attr->name);
g_kev_dumped = 1;
} else {
printf("[kev] fopen failed\n");
}
fflush(stdout);
}
return 0;
}
// ---- 极简 base64 解码,给 TOK 命令用(文本里有空格换行,用 base64 最省事)----
static int b64val(char c) {
if (c >= 'A' && c <= 'Z') return c - 'A';
if (c >= 'a' && c <= 'z') return c - 'a' + 26;
if (c >= '0' && c <= '9') return c - '0' + 52;
if (c == '+') return 62;
if (c == '/') return 63;
return -1;
}
static int b64decode(const char* in, int inlen, char* out, int outmax) {
int v = 0, bits = 0, n = 0;
for (int i = 0; i < inlen; i++) {
if (in[i] == '=' || in[i] == '\n' || in[i] == '\r') continue;
int d = b64val(in[i]);
if (d < 0) continue;
v = (v << 6) | d;
bits += 6;
if (bits >= 8) {
bits -= 8;
if (n < outmax) out[n++] = (char)((v >> bits) & 0xFF);
}
}
return n;
}
int main(int argc, char** argv)
{
// keep_history / ignore_eos_token / key_path are optional and parsed by
// argc position: passing fewer falls back to the defaults in [] below.
// Usage: %s <rknn> <weight> <tokenizer> <embedding> <max_context_len> <max_new_tokens> <core_mask> [keep_history] [ignore_eos_token]
// [key_path]
if (argc < 8 || argc > 12) { // 12 = 多一个 ids 文件
LOGW(
"Usage: %s <rknn_path> <weight_path> <tokenizer.gguf> <embedding.bin> <max_context_len> <max_new_tokens> <core_mask> [keep_history] [ignore_eos_token] [key_path]\n ",
argv[0]);
LOGW(
"Such as: %s ./model/Qwen2.5-0.5B.rknn ./model/Qwen2.5-0.5B.weight ./model/Qwen2.5-0.5B.tokenizer.gguf ./model/Qwen2.5-0.5B-embedding.bin 1024 256 0xff 0 0 ./model/key.env\n",
argv[0]);
LOGW(
"For encrypted model: %s ./model/Qwen2.5-0.5B.rknn ./model/Qwen2.5-0.5B.weight ./model/Qwen2.5-0.5B.tokenizer.gguf ./model/Qwen2.5-0.5B-embedding.bin 1024 256 0xff 0 0 ./model/key.env\n",
argv[0]);
return -1;
}
int ret = 0;
int i = 0; // 已处理的序列数(放在这里避免 goto 跨初始化)
char* model_path = argv[1];
char* weight_path = argv[2];
const char* tokenizer_path = argv[3];
const char* embedding_path = argv[4];
int32_t max_context_len = atoi(argv[5]);
int32_t max_new_tokens = atoi(argv[6]);
uint32_t core_mask = strtoul(argv[7], nullptr, 16);
// Optional positional args; defaults keep the old behaviour.
int keep_history = (argc > 8) ? atoi(argv[8]) : 0;
bool ignore_eos_token = (argc > 9) ? (atoi(argv[9]) != 0) : false;
const char* key_path = (argc > 10) ? argv[10] : nullptr;
const char* ids_path = (argc > 11) ? argv[11] : nullptr; // 每行一串 token id
// key 文件是占位参数;不存在就当没给,否则会被当成加密模型去解密
if (key_path) {
FILE* kf = fopen(key_path, "r");
if (kf) fclose(kf);
else {
printf("[kev] key file '%s' not found, ignore it\n", key_path);
key_path = nullptr;
}
}
rknn3_devices devs;
rknn3_context ctx = 0;
rknn3_input_output_num io_num;
VocabInfo vocab_info;
Tokenizer* tokenizer = nullptr;
struct stat emb_st;
struct embedding_info embedding_info;
RKLLMCallback callback;
rknn3_test_session_userdata_t session_userdata;
rknn3_llm_param params;
int n_params = 1;
RKLLMRunState state;
rknn3_lora loras_enabled[RKNN3_MAX_LORA_NUM];
rknn3_llm_config llm_config;
rknn3_kvcache_policy kvcache_policy = RKNN3_KVCACHE_POLICY_NORMAL;
rknn3_kvcache_policy_param kvcache_policy_param;
int llm_n_inputs = 1;
rknn3_llm_input* llm_inputs[llm_n_inputs];
rknn3_llm_infer_param llm_infer_param;
memset(&devs, 0, sizeof(devs));
memset(&io_num, 0, sizeof(io_num));
memset(&vocab_info, 0, sizeof(vocab_info));
memset(&emb_st, 0, sizeof(emb_st));
memset(&embedding_info, 0, sizeof(embedding_info));
memset(&callback, 0, sizeof(callback));
memset(¶ms, 0, sizeof(params));
memset(&state, 0, sizeof(state));
memset(&llm_config, 0, sizeof(llm_config));
memset(loras_enabled, 0, sizeof(loras_enabled));
memset(&kvcache_policy_param, 0, sizeof(kvcache_policy_param));
state.loras_enabled = loras_enabled;
memset(llm_inputs, 0, sizeof(llm_inputs));
memset(&llm_infer_param, 0, sizeof(llm_infer_param));
memset(&session_userdata, 0, sizeof(session_userdata));
// set output tensors index for output callback
rknn3_tensor output_tensors[1];
int n_output_tensors = 1;
int output_tensors_index[1] = {0};
memset(output_tensors, 0, sizeof(output_tensors));
std::vector<std::string> random_prompts = {
"请解释一下相对论的基本概念。",
"Please explain the basic concept of relativity",
};
int test_num = random_prompts.size();
// 若给了 ids 文件,就走批量模式:每行一串 token id,逐条 prefill 并 dump logits
std::vector<std::vector<int32_t>> seqs;
if (ids_path) {
FILE* fp = fopen(ids_path, "r");
if (!fp) { printf("cannot open ids file: %s\n", ids_path); return -1; }
char line[65536];
while (fgets(line, sizeof(line), fp)) {
std::vector<int32_t> one;
char* p = line;
while (*p) {
char* end = nullptr;
long v = strtol(p, &end, 10);
if (end == p) break;
one.push_back((int32_t)v);
p = end;
}
if (!one.empty()) seqs.push_back(one);
}
fclose(fp);
test_num = (int)seqs.size();
printf("[kev] loaded %d sequences from %s\n", test_num, ids_path);
}
printf("*******************************NEW TEST**********************************\n");
// find devices (devs already zero-initialised above)
const char* device_id = nullptr;
ret = rknn3_find_devices(&devs);
if (ret != RKNN3_SUCCESS || devs.n_devices == 0) {
printf("rknn3_find_devices failed, ret = %d\n", ret);
goto exit;
}
// init context and model
device_id = devs.devices[0].id;
ret = init_context_and_model(&ctx, model_path, weight_path, core_mask, key_path, device_id);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_init_model fail! ret = %d\n", ret);
goto exit;
}
// Get Model Output Info for output callback (if needed)
if (n_output_tensors > 0) {
ret = setup_output_tensors(ctx, output_tensors, n_output_tensors, output_tensors_index, &io_num);
if (ret < 0) {
goto exit;
}
}
// get tokenizer and embedding
ret = get_tokenizer_and_embedding(tokenizer_path, &vocab_info, &tokenizer, &embedding_info, embedding_path, &emb_st);
if (ret < 0) {
printf("get_tokenizer_and_embedding failed, ret = %d\n", ret);
goto exit;
}
ret = rknn3_query(ctx, RKNN3_QUERY_LLM_CONFIG, &llm_config, sizeof(rknn3_llm_config));
if (ret != RKNN3_SUCCESS) {
printf("rknn3_query llm config failed! ret=%d", ret);
goto exit;
}
// Set basic session parameters
build_default_llm_params(¶ms, &vocab_info, max_context_len, ignore_eos_token);
// init session
session_userdata.session = rknn3_session_init(ctx, ¶ms, n_params);
if (!session_userdata.session) {
printf("rknn3_session_init failed, ret = %d\n", ret);
goto exit;
}
// set session callback
build_default_callback(&callback, tokenizer, &embedding_info, &session_userdata);
// sampling_callback 不会被运行时调用(内部采样),logits 从 output_callback 拿
callback.sampling_callback = NULL;
callback.sampling_userdata = NULL;
// 同时打开 output_callback:若它能触发而 sampling 不能,说明 sampling 被门控
callback.output_callback = kev_output_callback;
callback.output_userdata = &embedding_info;
callback.output_tensors = output_tensors;
callback.n_output_tensors = n_output_tensors;
ret = rknn3_session_set_callback(session_userdata.session, &callback);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_set_callback failed, ret = %d\n", ret);
goto exit;
}
// set/update chat template
if (0) {
std::string system_prompt = "<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n";
std::string prompt_prefix = "<|im_start|>user\n";
std::string prompt_postfix = "<|im_end|>\n<|im_start|>assistant\n";
ret = rknn3_session_set_chat_template(session_userdata.session, system_prompt.c_str(), prompt_prefix.c_str(), prompt_postfix.c_str());
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_set_chat_template failed, ret = %d\n", ret);
goto exit;
}
}
// set kvcache policy
ret = rknn3_session_set_kvcache_policy(session_userdata.session, kvcache_policy, nullptr);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_set_kvcache_policy failed, ret=%d\n", ret);
goto exit;
}
print_llm_config(&llm_config);
// Test the generic session inference: each iteration builds a typical prompt request,
// runs one round of generation to show the standard prompt input workflow.
char in_line[65536];
// 关键:stdout 接管道时 C 运行库默认全缓冲,READY_DAEMON 会卡在缓冲区里,
// 父进程永远等不到。必须显式改成行缓冲(只 fflush 一次不够,后面每行都要)。
setvbuf(stdout, NULL, _IOLBF, 0);
printf("READY_DAEMON\n");
fflush(stdout);
while (fgets(in_line, sizeof(in_line), stdin)) {
if (strncmp(in_line, "quit", 4) == 0) break;
// OUT <path> -> 指定下一次 dump 的落盘路径
if (strncmp(in_line, "OUT ", 4) == 0) {
char* q = in_line + 4;
while (*q == ' ') q++;
int n = (int)strlen(q);
while (n > 0 && (q[n-1] == '\n' || q[n-1] == '\r')) n--;
if (n > 0 && n < (int)sizeof(g_out_path)) {
memcpy(g_out_path, q, n);
g_out_path[n] = 0;
printf("OUTSET %s\n", g_out_path);
fflush(stdout);
}
continue;
}
// TOK <base64 text> -> 打印 "IDS n id1 id2 ..."
if (strncmp(in_line, "TOK ", 4) == 0) {
static char tbuf[65536];
static int32_t tids[8192];
int raw = b64decode(in_line + 4, (int)strlen(in_line + 4), tbuf, sizeof(tbuf) - 1);
tbuf[raw] = 0;
printf("[kev] TOK raw=%d head='%.40s' tok=%p\n", raw, tbuf, (void*)tokenizer);
fflush(stdout);
int n = tokenizer->Tokenize(tbuf, raw, tids, 8192);
printf("IDS %d", n);
for (int k = 0; k < n; k++) printf(" %d", tids[k]);
printf("\n");
fflush(stdout);
continue;
}
// R <ids> = 清空 kvcache 后从头 prefill(keep_history=0)
// A <ids> = 接在上一次之后追加 prefill(keep_history=1)
// 不带前缀 = 老行为(等价 R)
bool append = false;
char* pp = in_line;
if (in_line[0] == 'A' && in_line[1] == ' ') { append = true; pp = in_line + 2; }
else if (in_line[0] == 'R' && in_line[1] == ' ') { pp = in_line + 2; }
{
std::vector<int32_t> one;
char* p = pp;
while (*p) {
char* e = nullptr;
long v = strtol(p, &e, 10);
if (e == p) break;
one.push_back((int32_t)v);
p = e;
}
if (one.empty()) continue;
seqs.push_back(one);
}
printf("\n--------------------Input[%d]-------------------- \n", i);
llm_infer_param = build_default_infer_param(max_new_tokens, append ? 1 : 0);
g_kev_seq = i;
g_kev_dumped = 0;
// A 模式接着上次的 kvcache;R 模式(默认)从头开始
if (!append) {
rknn3_session_clear_kvcache(session_userdata.session, RKNN3_KVCACHE_CLEAR_ALL);
}
llm_inputs[0] = build_llm_input_tokens(seqs.back().data(), seqs.back().size(), false, false);
printf("\n--------------------Output---------------------- \n");
rknn3_test_timing_mark_start(&session_userdata.timing);
ret = rknn3_session_run(session_userdata.session, llm_inputs[0], llm_n_inputs, &llm_infer_param);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_run failed, ret = %d\n", ret);
goto exit;
}
rknn3_test_timing_mark_end(&session_userdata.timing);
printf("DONE %d\n", i); // 每处理完一条就给客户端一个信号
fflush(stdout);
i++;
ret = rknn3_session_query_state(session_userdata.session, &state);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_query_state failed, ret=%d\n", ret);
goto exit;
}
print_run_state(&state, true);
// clear kvcache
if (state.n_total_tokens >= (state.n_max_tokens - max_new_tokens)) {
ret = rknn3_session_clear_kvcache(session_userdata.session, RKNN3_KVCACHE_CLEAR_ALL);
if (ret != RKNN3_SUCCESS) {
printf("rknn3_session_clear_kvcache failed, ret=%d\n", ret);
goto exit;
}
}
print_perf_stats(&state, &session_userdata.timing);
// Free any heap memory owned by this iteration's input(s).
for (int j = 0; j < llm_n_inputs; j++) {
release_llm_input(ctx, llm_inputs[j]);
}
}
exit:
for (int j = 0; j < llm_n_inputs; j++) {
release_llm_input(ctx, llm_inputs[j]);
}
if (n_output_tensors > 0) {
cleanup_output_tensors(ctx, output_tensors, n_output_tensors);
}
if (session_userdata.session) {
rknn3_session_destroy(session_userdata.session);
}
if (ctx) {
rknn3_destroy(ctx);
}
embedding_info_release(&embedding_info, &emb_st);
if (tokenizer) {
delete tokenizer;
}
printf("*******************************END TEST**********************************\n");
fflush(stdout);
return 0;
}
|