vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-13 02:19:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	Makefile
#	README.md
#	flake.lock
#	ggml-cuda.cu
#	llama.cpp
#	tests/test-backend-ops.cpp
#	tests/test-quantize-fns.cpp
This commit is contained in:
Concedo 2024-02-28 13:41:35 +08:00
commit ad638285de
26 changed files with 3393 additions and 589 deletions
Download patch file Download diff file Expand all files Collapse all files

4
examples/infill/infill.cpp
View file

@ -448,8 +448,8 @@ int main(int argc, char ** argv) {
LOG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
n_past, n_left, n_ctx, params.n_keep, n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep + 1 , params.n_keep + n_discard + 1);
llama_kv_cache_seq_shift(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep + 1 , params.n_keep + n_discard + 1);
llama_kv_cache_seq_add(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
n_past -= n_discard;

10
examples/main/main.cpp
View file

@ -549,8 +549,8 @@ int main(int argc, char ** argv) {
LOG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
n_past, n_left, n_ctx, params.n_keep, n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_cache_seq_shift(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
llama_kv_cache_seq_rm (ctx, 0, params.n_keep , params.n_keep + n_discard);
llama_kv_cache_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
n_past -= n_discard;
@ -577,9 +577,9 @@ int main(int argc, char ** argv) {
LOG("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n, (ga_i + ib*bd)/ga_n, (ga_i + ib*bd + ga_w)/ga_n);
LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib*bd + ga_w, n_past + ib*bd, dd, ga_i + ib*bd + ga_w + dd, n_past + ib*bd + dd);
llama_kv_cache_seq_shift(ctx, 0, ga_i, n_past, ib*bd);
llama_kv_cache_seq_div (ctx, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_kv_cache_seq_shift(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
llama_kv_cache_seq_add(ctx, 0, ga_i, n_past, ib*bd);
llama_kv_cache_seq_div(ctx, 0, ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n);
llama_kv_cache_seq_add(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd, dd);
n_past -= bd;

25
examples/passkey/passkey.cpp
View file

@ -126,7 +126,7 @@ int main(int argc, char ** argv) {
const int n_batch = ctx_params.n_batch;
const int n_batch_grp = ctx_params.n_batch/n_grp;
LOG_TEE("\n%s: n_len = %d, n_ctx = %d, n_kv_req = %d, n_grp = %d, n_batch = %d\n", __func__, n_len, n_ctx, n_kv_req, n_grp, n_batch);
LOG_TEE("\n%s: n_len = %d, n_ctx = %d, n_kv_req = %d, n_grp = %d, n_batch = %d, n_junk = %d, i_pos = %d\n", __func__, n_len, n_ctx, n_kv_req, n_grp, n_batch, n_junk, i_pos);
// print the prompt token-by-token
@ -146,10 +146,11 @@ int main(int argc, char ** argv) {
const int ib = i/n_batch - 1;
const int bd = n_batch_grp*(n_grp - 1);
llama_kv_cache_seq_shift(ctx, 0, n_past - n_batch, n_past, ib*bd);
llama_kv_cache_seq_div (ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
llama_kv_cache_seq_add (ctx, 0, n_past - n_batch, n_past, ib*bd);
llama_kv_cache_seq_div (ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
llama_kv_cache_update (ctx);
n_past -= bd;
n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
}
llama_batch_clear(batch);
@ -179,10 +180,12 @@ int main(int argc, char ** argv) {
LOG_TEE("%s: shifting KV cache with %d\n", __func__, n_discard);
llama_kv_cache_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_cache_seq_shift(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
llama_kv_cache_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_cache_seq_add(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
//llama_kv_cache_defrag (ctx);
llama_kv_cache_update (ctx);
n_past -= n_discard;
n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
llama_batch_clear(batch);
@ -208,10 +211,12 @@ int main(int argc, char ** argv) {
if (n_discard > 0) {
LOG_TEE("%s: shifting KV cache with %d to free space for the answer\n", __func__, n_discard);
llama_kv_cache_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_cache_seq_shift(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
llama_kv_cache_seq_rm (ctx, 0, n_keep , n_keep + n_discard);
llama_kv_cache_seq_add(ctx, 0, n_keep + n_discard, n_ctx, -n_discard);
//llama_kv_cache_defrag (ctx);
llama_kv_cache_update (ctx);
n_past -= n_discard;
n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
}
}

10
examples/quantize/quantize.cpp
View file

@ -24,18 +24,21 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
{ "Q5_1", LLAMA_FTYPE_MOSTLY_Q5_1, " 4.70G, +0.0349 ppl @ LLaMA-v1-7B", },
{ "IQ2_XXS",LLAMA_FTYPE_MOSTLY_IQ2_XXS," 2.06 bpw quantization", },
{ "IQ2_XS", LLAMA_FTYPE_MOSTLY_IQ2_XS, " 2.31 bpw quantization", },
{ "IQ2_S", LLAMA_FTYPE_MOSTLY_IQ2_S, " 2.5 bpw quantization", },
{ "IQ2_M", LLAMA_FTYPE_MOSTLY_IQ2_M, " 2.7 bpw quantization", },
{ "IQ1_S", LLAMA_FTYPE_MOSTLY_IQ1_S, " 1.56 bpw quantization", },
{ "Q2_K", LLAMA_FTYPE_MOSTLY_Q2_K, " 2.63G, +0.6717 ppl @ LLaMA-v1-7B", },
{ "Q2_K_S", LLAMA_FTYPE_MOSTLY_Q2_K_S, " 2.16G, +9.0634 ppl @ LLaMA-v1-7B", },
{ "IQ3_XXS",LLAMA_FTYPE_MOSTLY_IQ3_XXS," 3.06 bpw quantization", },
{ "IQ3_S", LLAMA_FTYPE_MOSTLY_IQ3_S, " 3.44 bpw quantization", },
{ "IQ3_M", LLAMA_FTYPE_MOSTLY_IQ3_M, " 3.66 bpw quantization mix", },
{ "IQ3_M", LLAMA_FTYPE_MOSTLY_IQ3_M, " 3.66 bpw quantization mix", },
{ "Q3_K", LLAMA_FTYPE_MOSTLY_Q3_K_M, "alias for Q3_K_M" },
{ "Q3_K_XS",LLAMA_FTYPE_MOSTLY_Q3_K_XS,"3-bit extra small quantization" , },
{ "IQ3_XS", LLAMA_FTYPE_MOSTLY_IQ3_XS, " 3.3 bpw quantization" , },
{ "Q3_K_S", LLAMA_FTYPE_MOSTLY_Q3_K_S, " 2.75G, +0.5551 ppl @ LLaMA-v1-7B", },
{ "Q3_K_M", LLAMA_FTYPE_MOSTLY_Q3_K_M, " 3.07G, +0.2496 ppl @ LLaMA-v1-7B", },
{ "Q3_K_L", LLAMA_FTYPE_MOSTLY_Q3_K_L, " 3.35G, +0.1764 ppl @ LLaMA-v1-7B", },
{ "IQ4_NL", LLAMA_FTYPE_MOSTLY_IQ4_NL, " 4.25 bpw non-linear quantization", },
{ "IQ4_NL", LLAMA_FTYPE_MOSTLY_IQ4_NL, " 4.50 bpw non-linear quantization", },
{ "IQ4_XS", LLAMA_FTYPE_MOSTLY_IQ4_XS, " 4.25 bpw non-linear quantization", },
{ "Q4_K", LLAMA_FTYPE_MOSTLY_Q4_K_M, "alias for Q4_K_M", },
{ "Q4_K_S", LLAMA_FTYPE_MOSTLY_Q4_K_S, " 3.59G, +0.0992 ppl @ LLaMA-v1-7B", },
{ "Q4_K_M", LLAMA_FTYPE_MOSTLY_Q4_K_M, " 3.80G, +0.0532 ppl @ LLaMA-v1-7B", },
@ -293,6 +296,7 @@ int main(int argc, char ** argv) {
}
if ((params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS ||
params.ftype == LLAMA_FTYPE_MOSTLY_IQ2_S ||
params.ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S || params.ftype == LLAMA_FTYPE_MOSTLY_IQ1_S) && imatrix_data.empty()) {
fprintf(stderr, "\n===============================================================================================\n");
fprintf(stderr, "Please do not use IQ1_S, IQ2_XXS, IQ2_XS or Q2_K_S quantization without an importance matrix\n");

18
examples/server/README.md
View file

@ -1,8 +1,20 @@
# llama.cpp/example/server
# LLaMA.cpp HTTP Server
This example demonstrates a simple HTTP API server and a simple web front end to interact with llama.cpp.
Fast, lightweight, pure C/C++ HTTP server based on [httplib](https://github.com/yhirose/cpp-httplib), [nlohmann::json](https://github.com/nlohmann/json) and **llama.cpp**.
Command line options:
Set of LLM REST APIs and a simple web front end to interact with llama.cpp.
**Features:**
* LLM inference of F16 and quantum models on GPU and CPU
* [OpenAI API](https://github.com/openai/openai-openapi) compatible chat completions and embeddings routes
* Parallel decoding with multi-user support
* Continuous batching
* Multimodal (wip)
* Monitoring endpoints
The project is under active development, and we are [looking for feedback and contributors](https://github.com/ggerganov/llama.cpp/issues/4216).
**Command line options:**
- `--threads N`, `-t N`: Set the number of threads to use during generation.
- `-tb N, --threads-batch N`: Set the number of threads to use during batch and prompt processing. If not specified, the number of threads will be set to the number of threads used for generation.

12
examples/server/server.cpp
View file

@ -1337,6 +1337,10 @@ struct llama_server_context
split_multiprompt_task(task_id, task);
}
} else {
// an empty prompt can make slot become buggy
if (task.data.contains("prompt") && task.data["prompt"].is_string() && task.data["prompt"].get<std::string>().empty()) {
task.data["prompt"] = " "; // add a space so that we have one token
}
queue_tasks.post(task);
}
}
@ -1637,8 +1641,8 @@ struct llama_server_context
{"n_system_tokens", system_tokens.size()},
{"n_cache_tokens", slot.cache_tokens.size()}
});
llama_kv_cache_seq_rm (ctx, slot.id, n_keep , n_keep + n_discard);
llama_kv_cache_seq_shift(ctx, slot.id, n_keep + n_discard, system_tokens.size() + slot.n_past, -n_discard);
llama_kv_cache_seq_rm (ctx, slot.id, n_keep , n_keep + n_discard);
llama_kv_cache_seq_add(ctx, slot.id, n_keep + n_discard, system_tokens.size() + slot.n_past, -n_discard);
for (size_t i = n_keep + n_discard; i < slot.cache_tokens.size(); i++)
{
@ -1942,9 +1946,9 @@ struct llama_server_context
LOG_TEE("div: [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
llama_kv_cache_seq_shift(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
llama_kv_cache_seq_div(ctx, slot.id, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w,slot.ga_n);
llama_kv_cache_seq_shift(ctx, slot.id, slot.ga_i + ib * bd + slot.ga_w,slot.n_past_se + ib * bd, dd);
llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i + ib * bd + slot.ga_w,slot.n_past_se + ib * bd, dd);
slot.n_past_se -= bd;

4
examples/server/tests/features/steps/steps.py
View file

@ -699,6 +699,8 @@ async def wait_for_health_status(context,
if context.debug:
print(f"Starting checking for health for expected_health_status={expected_health_status}")
timeout = 3 # seconds
if expected_health_status == 'ok':
timeout = 10 # CI slow inference
interval = 0.5
counter = 0
async with aiohttp.ClientSession() as session:
@ -736,7 +738,7 @@ async def wait_for_health_status(context,
if n_completions > 0:
return
assert False, 'timeout exceeded'
assert False, f'{expected_health_status} timeout exceeded {counter}s>={timeout}'
def assert_embeddings(embeddings):