Merge pull request #4987 from ollama/mxyng/revert-byte-order

Revert "Merge pull request #4938 from ollama/mxyng/fix-byte-order"

README.md

@@ -332,6 +332,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
 - [Ollama for Ruby](https://github.com/gbaptista/ollama-ai)
 - [Ollama-rs for Rust](https://github.com/pepperoni21/ollama-rs)
+- [Ollama-hpp for C++](https://github.com/jmont-dev/ollama-hpp)
 - [Ollama4j for Java](https://github.com/amithkoujalgi/ollama4j)
 - [ModelFusion Typescript Library](https://modelfusion.dev/integration/model-provider/ollama)
 - [OllamaKit for Swift](https://github.com/kevinhermawan/OllamaKit)

docs/api.md

@@ -250,7 +250,7 @@ curl http://localhost:11434/api/generate -d '{
 
 #### Request (Reproducible outputs)
 
-For reproducible outputs, set `temperature` to 0 and `seed` to a number:
+For reproducible outputs, set `seed` to a number:
 
 ##### Request
 
@@ -259,8 +259,7 @@ curl http://localhost:11434/api/generate -d '{
   "model": "mistral",
   "prompt": "Why is the sky blue?",
   "options": {
-    "seed": 123,
-    "temperature": 0
+    "seed": 123
   }
 }'
 ```

llm/ext_server/server.cpp

@@ -359,7 +359,6 @@ struct llama_server_context
 
     // slots / clients
     std::vector<server_slot> slots;
-    json default_generation_settings_for_props;
 
     llama_server_queue    queue_tasks;
     llama_server_response queue_results;
@@ -483,9 +482,6 @@ struct llama_server_context
             slots.push_back(slot);
         }
 
-        default_generation_settings_for_props = get_formated_generation(slots.front());
-        default_generation_settings_for_props["seed"] = -1;
-
         batch = llama_batch_init(n_ctx, 0, params.n_parallel);
     }
 
@@ -584,7 +580,7 @@ struct llama_server_context
         slot->sparams.mirostat_eta      = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
         slot->sparams.penalize_nl       = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
         slot->params.n_keep             = json_value(data, "n_keep",            slot->params.n_keep);
-        slot->params.seed               = json_value(data, "seed",              default_params.seed);
+        slot->sparams.seed              = json_value(data, "seed",              default_params.seed);
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
@@ -811,7 +807,6 @@ struct llama_server_context
             llama_sampling_free(slot->ctx_sampling);
         }
         slot->ctx_sampling = llama_sampling_init(slot->sparams);
-        llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
         all_slots_are_idle = false;

llm/ggml.go

@@ -231,7 +231,8 @@ const (
 	// Magic constant for `ggla` files (LoRA adapter).
 	FILE_MAGIC_GGLA = 0x67676C61
 	// Magic constant for `gguf` files (versioned, gguf)
-	FILE_MAGIC_GGUF = 0x46554747
+	FILE_MAGIC_GGUF_LE = 0x46554747
+	FILE_MAGIC_GGUF_BE = 0x47475546
 )
 
 var ErrUnsupportedFormat = errors.New("unsupported model format")
@@ -246,7 +247,7 @@ func DetectGGMLType(b []byte) string {
 		return "ggjt"
 	case FILE_MAGIC_GGLA:
 		return "ggla"
-	case FILE_MAGIC_GGUF:
+	case FILE_MAGIC_GGUF_LE, FILE_MAGIC_GGUF_BE:
 		return "gguf"
 	default:
 		return ""
@@ -254,19 +255,21 @@ func DetectGGMLType(b []byte) string {
 }
 
 func DecodeGGML(rs io.ReadSeeker) (*GGML, int64, error) {
-	var magic [4]byte
+	var magic uint32
 	if err := binary.Read(rs, binary.LittleEndian, &magic); err != nil {
 		return nil, 0, err
 	}
 
 	var c container
-	switch binary.LittleEndian.Uint32(magic[:]) {
+	switch magic {
 	case FILE_MAGIC_GGML, FILE_MAGIC_GGMF, FILE_MAGIC_GGJT:
 		return nil, 0, ErrUnsupportedFormat
 	case FILE_MAGIC_GGLA:
 		c = &containerGGLA{}
-	case FILE_MAGIC_GGUF:
+	case FILE_MAGIC_GGUF_LE:
 		c = &containerGGUF{ByteOrder: binary.LittleEndian}
+	case FILE_MAGIC_GGUF_BE:
+		c = &containerGGUF{ByteOrder: binary.BigEndian}
 	default:
 		return nil, 0, errors.New("invalid file magic")
 	}

llm/gguf.go

@@ -36,23 +36,10 @@ func (c *containerGGUF) Name() string {
 }
 
 func (c *containerGGUF) Decode(rs io.ReadSeeker) (model, error) {
-	var version [4]byte
-	if err := binary.Read(rs, c.ByteOrder, &version); err != nil {
+	if err := binary.Read(rs, c.ByteOrder, &c.Version); err != nil {
 		return nil, err
 	}
 
-	// if the lower 16 bits are 0, the byte order is probably wrong
-	if c.ByteOrder.Uint32(version[:])&1<<4 == 0 {
-		switch c.ByteOrder {
-		case binary.LittleEndian:
-			c.ByteOrder = binary.BigEndian
-		case binary.BigEndian:
-			c.ByteOrder = binary.LittleEndian
-		}
-	}
-
-	c.Version = c.ByteOrder.Uint32(version[:])
-
 	var err error
 	switch c.Version {
 	case 1: