talk-llama : sync llama.cpp

This commit is contained in:
Georgi Gerganov
2026-01-30 15:56:40 +02:00
parent b529c0610f
commit 953e503fd9
38 changed files with 2299 additions and 730 deletions
+13 -6
View File
@@ -244,11 +244,14 @@ struct llama_file::impl {
}
errno = 0;
if (fd == -1) {
std::size_t ret = std::fread(ptr, len, 1, fp);
const size_t curr_off = tell();
const size_t to_read = std::min(len, size - curr_off);
std::size_t ret = std::fread(ptr, to_read, 1, fp);
if (ferror(fp)) {
throw std::runtime_error(format("read error: %s", strerror(errno)));
}
if (ret != 1) {
if (to_read > 0 && ret != 1) {
throw std::runtime_error("unexpectedly reached end of file");
}
} else {
@@ -262,7 +265,8 @@ struct llama_file::impl {
continue; // Interrupted by signal, retry
}
// Fallback to std::fread in case the DMA controller cannot access the buffer
if (errno == EFAULT) {
if (errno == EFAULT || errno == EINVAL) {
LLAMA_LOG_WARN("%s: Falling back to buffered IO due to %s\n", __func__, strerror(errno));
auto curr_off = tell();
close(fd);
fd = -1;
@@ -381,6 +385,9 @@ int llama_file::file_id() const {
#ifdef _WIN32
return _fileno(pimpl->fp);
#else
if (pimpl->fd != -1) {
return pimpl->fd;
}
#if defined(fileno)
return fileno(pimpl->fp);
#else
@@ -611,9 +618,9 @@ struct llama_mlock::impl {
char* errmsg = std::strerror(errno);
bool suggest = (errno == ENOMEM);
#if defined(TARGET_OS_VISION) || defined(TARGET_OS_TV) || defined(_AIX)
// visionOS/tvOS dont't support RLIMIT_MEMLOCK
// Skip resource limit checks on visionOS/tvOS
#if defined(TARGET_OS_VISION) || defined(TARGET_OS_TV) || defined(_AIX) || defined(__HAIKU__)
// visionOS/tvOS/Haiku don't support RLIMIT_MEMLOCK
// Skip resource limit checks on these platforms
suggest = false;
#else
struct rlimit lock_limit;