common : remove chatml and instruct params

ggml-ci

common/common.cpp

@@ -238,10 +238,7 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
         }
     }
 
-    if (params.prompt_cache_all &&
+    if (params.prompt_cache_all && (params.interactive || params.interactive_first)) {
-            (params.interactive || params.interactive_first ||
-             params.instruct)) {
-
         throw std::invalid_argument("error: --prompt-cache-all not supported in interactive mode yet\n");
     }
 
@@ -920,18 +917,10 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
         params.interactive_first = true;
         return true;
     }
-    if (arg == "-ins" || arg == "--instruct") {
-        params.instruct = true;
-        return true;
-    }
     if (arg == "-cnv" || arg == "--conversation") {
         params.conversation = true;
         return true;
     }
-    if (arg == "-cml" || arg == "--chatml") {
-        params.chatml = true;
-        return true;
-    }
     if (arg == "--infill") {
         params.infill = true;
         return true;
@@ -1645,8 +1634,6 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
                                                                         "can be specified more than once for multiple prompts" });
     options.push_back({ "main",        "-sp,   --special",              "special tokens output enabled (default: %s)", params.special ? "true" : "false" });
     options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode (does not print special tokens and suffix/prefix) (default: %s)", params.conversation ? "true" : "false" });
-    options.push_back({ "main",        "-ins,  --instruct",             "run in instruction mode (use with Alpaca models) (default: %s)", params.instruct ? "true" : "false" });
-    options.push_back({ "main",        "-cml,  --chatml",               "run in chatml mode (use with ChatML-compatible models) (default: %s)", params.chatml ? "true" : "false" });
     options.push_back({ "main infill", "-i,    --interactive",          "run in interactive mode (default: %s)", params.interactive ? "true" : "false" });
     options.push_back({ "main infill", "-if,   --interactive-first",    "run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false" });
     options.push_back({ "main infill", "-mli,  --multiline-input",      "allows you to write or paste multiple lines without ending each in '\\'" });
@@ -3200,7 +3187,6 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     yaml_dump_string_multiline(stream, "in_prefix", params.input_prefix.c_str());
     fprintf(stream, "in_prefix_bos: %s # default: false\n", params.input_prefix_bos ? "true" : "false");
     yaml_dump_string_multiline(stream, "in_suffix", params.input_prefix.c_str());
-    fprintf(stream, "instruct: %s # default: false\n", params.instruct ? "true" : "false");
     fprintf(stream, "interactive: %s # default: false\n", params.interactive ? "true" : "false");
     fprintf(stream, "interactive_first: %s # default: false\n", params.interactive_first ? "true" : "false");
     fprintf(stream, "keep: %d # default: 0\n", params.n_keep);

common/common.h

@@ -148,7 +148,6 @@ struct gpt_params {
     bool interactive       = false; // interactive mode
     bool interactive_first = false; // wait for user input immediately
     bool conversation      = false; // conversation mode (does not print special tokens and suffix/prefix)
-    bool chatml            = false; // chatml mode (used for models trained on chatml syntax)
     bool prompt_cache_all  = false; // save user input and generations to prompt cache
     bool prompt_cache_ro   = false; // open the prompt cache read-only and do not update it
 
@@ -161,7 +160,6 @@ struct gpt_params {
 
     bool input_prefix_bos  = false; // prefix BOS to user inputs, preceding input_prefix
     bool ignore_eos        = false; // ignore generated EOS tokens
-    bool instruct          = false; // instruction mode (used for Alpaca models)
     bool logits_all        = false; // return logits for all tokens in the batch
     bool use_mmap          = true;  // use mmap for faster loads
     bool use_mlock         = false; // use mlock to keep model in memory

examples/main/main.cpp

@@ -249,11 +249,8 @@ int main(int argc, char ** argv) {
 
     std::vector<llama_token> embd_inp;
 
-    if (params.interactive_first || params.instruct || params.chatml || !params.prompt.empty() || session_tokens.empty()) {
+    if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) {
         LOG("tokenize the prompt\n");
-        if (params.chatml) {
-            params.prompt = "<|im_start|>system\n" + params.prompt + "<|im_end|>";
-        }
         embd_inp = ::llama_tokenize(ctx, params.prompt, true, true);
     } else {
         LOG("use session tokens\n");
@@ -331,37 +328,13 @@ int main(int argc, char ** argv) {
     }
 
     // number of tokens to keep when resetting context
-    if (params.n_keep < 0 || params.n_keep > (int) embd_inp.size() || params.instruct || params.chatml) {
+    if (params.n_keep < 0 || params.n_keep > (int) embd_inp.size()) {
         params.n_keep = (int)embd_inp.size();
     } else {
         params.n_keep += add_bos; // always keep the BOS token
     }
 
-    // prefix & suffix for instruct mode
+    if (params.conversation) {
-    const auto inp_pfx = ::llama_tokenize(ctx, "\n\n### Instruction:\n\n", true,  true);
-    const auto inp_sfx = ::llama_tokenize(ctx, "\n\n### Response:\n\n",    false, true);
-
-    LOG("inp_pfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, inp_pfx).c_str());
-    LOG("inp_sfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, inp_sfx).c_str());
-
-    // chatml prefix & suffix
-    const auto cml_pfx = ::llama_tokenize(ctx, "\n<|im_start|>user\n", true, true);
-    const auto cml_sfx = ::llama_tokenize(ctx, "<|im_end|>\n<|im_start|>assistant\n", false, true);
-
-    LOG("cml_pfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, cml_pfx).c_str());
-    LOG("cml_sfx: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, cml_sfx).c_str());
-
-    // in instruct mode, we inject a prefix and a suffix to each input by the user
-    if (params.instruct) {
-        params.interactive_first = true;
-        params.antiprompt.emplace_back("### Instruction:\n\n");
-    }
-    // similar for chatml mode
-    else if (params.chatml) {
-        params.interactive_first = true;
-        params.antiprompt.emplace_back("<|im_start|>user\n");
-    }
-    else if (params.conversation) {
         params.interactive_first = true;
     }
 
@@ -822,15 +795,13 @@ int main(int argc, char ** argv) {
 
                     is_interacting = true;
                     printf("\n");
-                } else if (params.instruct || params.chatml) {
-                    is_interacting = true;
                 }
             }
 
             if (n_past > 0 && is_interacting) {
                 LOG("waiting for user input\n");
 
-                if (params.conversation || params.instruct || params.chatml) {
+                if (params.conversation) {
                     printf("\n> ");
                 }
 
@@ -873,18 +844,6 @@ int main(int argc, char ** argv) {
 
                     const size_t original_size = embd_inp.size();
 
-                    // instruct mode: insert instruction prefix
-                    if (params.instruct && !is_antiprompt) {
-                        LOG("inserting instruction prefix\n");
-                        n_consumed = embd_inp.size();
-                        embd_inp.insert(embd_inp.end(), inp_pfx.begin(), inp_pfx.end());
-                    }
-                    // chatml mode: insert user chat prefix
-                    if (params.chatml && !is_antiprompt) {
-                        LOG("inserting chatml prefix\n");
-                        n_consumed = embd_inp.size();
-                        embd_inp.insert(embd_inp.end(), cml_pfx.begin(), cml_pfx.end());
-                    }
                     if (params.escape) {
                         string_process_escapes(buffer);
                     }
@@ -899,17 +858,6 @@ int main(int argc, char ** argv) {
                     embd_inp.insert(embd_inp.end(), line_inp.begin(), line_inp.end());
                     embd_inp.insert(embd_inp.end(), line_sfx.begin(), line_sfx.end());
 
-                    // instruct mode: insert response suffix
-                    if (params.instruct) {
-                        LOG("inserting instruction suffix\n");
-                        embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());
-                    }
-                    // chatml mode: insert assistant chat suffix
-                    if (params.chatml) {
-                        LOG("inserting chatml suffix\n");
-                        embd_inp.insert(embd_inp.end(), cml_sfx.begin(), cml_sfx.end());
-                    }
-
                     for (size_t i = original_size; i < embd_inp.size(); ++i) {
                         const llama_token token = embd_inp[i];
                         output_tokens.push_back(token);
@@ -934,7 +882,7 @@ int main(int argc, char ** argv) {
         }
 
         // end of generation
-        if (!embd.empty() && llama_token_is_eog(model, embd.back()) && !(params.instruct || params.interactive || params.chatml)) {
+        if (!embd.empty() && llama_token_is_eog(model, embd.back()) && !(params.interactive)) {
             LOG_TEE(" [end of text]\n");
             break;
         }

scripts/run-with-preset.py

@@ -13,7 +13,7 @@ logger = logging.getLogger("run-with-preset")
 CLI_ARGS_MAIN_PERPLEXITY = [
     "batch-size", "cfg-negative-prompt", "cfg-scale", "chunks", "color", "ctx-size", "escape",
     "export", "file", "frequency-penalty", "grammar", "grammar-file", "hellaswag",
-    "hellaswag-tasks", "ignore-eos", "in-prefix", "in-prefix-bos", "in-suffix", "instruct",
+    "hellaswag-tasks", "ignore-eos", "in-prefix", "in-prefix-bos", "in-suffix",
     "interactive", "interactive-first", "keep", "logdir", "logit-bias", "lora", "lora-base",
     "low-vram", "main-gpu", "memory-f32", "mirostat", "mirostat-ent", "mirostat-lr", "mlock",
     "model", "multiline-input", "n-gpu-layers", "n-predict", "no-mmap", "no-mul-mat-q",