convert : auto-determine model name based on dir + scripts update

convert.py

@@ -106,6 +106,9 @@ class Params:
 
     ftype: Optional[GGMLFileType] = None
 
+    # path to the directory containing the model files
+    path_model: Optional['Path'] = None
+
     @staticmethod
     def find_n_mult(n_ff: int, n_embd: int) -> int:
         # hardcoded magic range
@@ -231,6 +234,8 @@ def load(model_plus: 'ModelPlus') -> 'Params':
         else:
             params = Params.guessed(model_plus.model)
 
+        params.path_model = model_plus.paths[0].parent
+
         return params
 
 
@@ -733,11 +738,13 @@ def __init__(self, fname_out: Path) -> None:
         self.gguf = gguf.GGUFWriter(fname_out, gguf.MODEL_ARCH_NAMES[ARCH])
 
     def add_meta_arch(self, params: Params) -> None:
-        ver = None
+        name = "LLaMA"
         if (params.n_ctx == 4096):
-            ver = "v2"
+            name = "LLaMA v2"
+            if params.path_model:
+                name = str(params.path_model.parent).split('/')[-1]
 
-        self.gguf.add_name                ("LLaMA" if ver == None else "LLaMA " + ver)
+        self.gguf.add_name                (name)
         self.gguf.add_context_length      (params.n_ctx)
         self.gguf.add_embedding_length    (params.n_embd)
         self.gguf.add_block_count         (params.n_layer)

scripts/qnt-all.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+qnt=(q8_0 q6_k q5_k q5_1 q5_0 q4_k q4_1 q4_0 q3_k q2_k)
+args=""
+
+if [ -z "$1" ]; then
+    echo "usage: $0 <model> [qnt] [args]"
+    echo "default: $0 <model> \"${qnt[@]}\" \"${args}\""
+    exit 1
+fi
+
+if [ ! -z "$2" ]; then
+    qnt=($2)
+fi
+
+if [ ! -z "$3" ]; then
+    args="$3"
+fi
+
+model="$1"
+out="../tmp/results-${model}"
+
+mkdir -p ${out}
+
+for q in ${qnt[@]}; do
+    time ./bin/quantize ../models/${model}/ggml-model-f16.gguf ../models/${model}/ggml-model-${q}.gguf ${q} 2>&1 ${args} | tee ${out}/qnt-${q}.txt
+done

scripts/run-all-perf.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+qnt=(f16 q8_0 q6_k q5_k q5_1 q5_0 q4_k q4_1 q4_0 q3_k q2_k)
+args="-ngl 999 -n 64 -p 512"
+
+if [ -z "$1" ]; then
+    echo "usage: $0 <model> [qnt] [args]"
+    echo "default: $0 <model> \"${qnt[@]}\" \"${args}\""
+    exit 1
+fi
+
+if [ ! -z "$2" ]; then
+    qnt=($2)
+fi
+
+if [ ! -z "$3" ]; then
+    args="$3"
+fi
+
+model="$1"
+out="../tmp/results-${model}"
+
+mkdir -p ${out}
+
+mstr=""
+
+for q in ${qnt[@]}; do
+    mstr="${mstr} -m ../models/${model}/ggml-model-${q}.gguf"
+done
+
+./bin/llama-bench ${mstr} ${args} 2> /dev/null

scripts/run-all-ppl.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+qnt=(f16 q8_0 q6_k q5_k q5_1 q5_0 q4_k q4_1 q4_0 q3_k q2_k)
+args="--no-mmap -ngl 999 -t 8"
+
+if [ -z "$1" ]; then
+    echo "usage: $0 <model> [qnt] [args]"
+    echo "default: $0 <model> \"${qnt[@]}\" \"${args}\""
+    exit 1
+fi
+
+if [ ! -z "$2" ]; then
+    qnt=($2)
+fi
+
+if [ ! -z "$3" ]; then
+    args="$3"
+fi
+
+model="$1"
+out="../tmp/results-${model}"
+
+mkdir -p ${out}
+
+for q in ${qnt[@]}; do
+    time ./bin/perplexity -m ../models/${model}/ggml-model-f16.gguf -f ./wiki.test.raw ${args} 2>&1 | tee ${out}/ppl-${q}.txt
+done