Merge pull request ggerganov#2 from jploski/master

Fixed bos/eos token (which is both 11 according to config.json of Fal…

examples/falcon/falcon_main.cpp

@@ -175,10 +175,8 @@ int main(int argc, char ** argv) {
     std::vector<llama_token> embd_inp;
 
     if (params.interactive_first || params.instruct || !params.prompt.empty() || session_tokens.empty()) {
-        // Add a space in front of the first character to match OG llama tokenizer behavior
-        params.prompt.insert(0, 1, ' ');
-
-        embd_inp = ::falcon_tokenize(ctx, params.prompt, true);
+        // Falcon does not have a dedicated bos token (bos==eos), so don't inject it here
+        embd_inp = ::falcon_tokenize(ctx, params.prompt, false);
     } else {
         embd_inp = session_tokens;
     }
@@ -361,9 +359,6 @@ int main(int argc, char ** argv) {
             if (n_past + (int) embd.size() > n_ctx) {
                 const int n_left = n_past - params.n_keep;
 
-                // always keep the first token - BOS
-                n_past = std::max(1, params.n_keep);
-
                 // insert n_left/2 tokens at the start of embd from last_n_tokens
                 embd.insert(embd.begin(), last_n_tokens.begin() + n_ctx - n_left/2 - embd.size(), last_n_tokens.end() - embd.size());
 

libfalcon.cpp

@@ -1348,12 +1348,6 @@ static bool falcon_eval_internal(
             const int    n_threads,
             const char * cgraph_fname) {
 
-    // enforce that the first token is BOS
-    if (n_past == 0 && tokens[0] != falcon_token_bos()) {
-        fprintf(stderr, "%s: first token must be BOS\n", __func__);
-        return false;
-    }
-
     const int64_t t_start_us = ggml_time_us();
 
     const int N = n_tokens;
@@ -3389,11 +3383,11 @@ const char * falcon_token_to_str(const struct falcon_context * ctx, llama_token
 }
 
 llama_token falcon_token_bos() {
-    return 1;
+    return 11;
 }
 
 llama_token falcon_token_eos() {
-    return 2;
+    return 11;
 }
 
 llama_token falcon_token_nl() {