llm : add Falcon support (#2717)

* llama : refactor GGUF constants into static maps

* llama : check if model architecture is known

* llama : refactor llama_model_load_internal()

* gguf : add KV constant maps

* llm : read arch-specific KVs

* convert : add dummy scores + types

* falcon : load tensor data (CPU only)

* llama : fix loading progress bar

* llama : add arch member to llama_model

* falcon : CPU inference working

* falcon : support non-40B models

* falcon : minor

* llama : minor updates

ggml-ci

* convert-falcon-hf-to-gguf.py : fix special token mapping

* llama.cpp : llama default UNK token = id 0

* llama.cpp : fix bpe tokenizer

* llama.cpp : fix the fix of bpe tokenizer

* ggml : pass eps to ggml_norm

* metal : implement RoPE (mode = 2) + avoid ggml_repeat

* ggml : ggml_repeat always creates new tensor

* falcon : copy-paste self-attention from LLaMA

* metal : print extra compute pipeline info

* falcon : minor changes (still chasing the Metal problem)

* llama.cpp : fix linefeed token

* metal : fix GELU kernel numerical stability by using precise::tanh

* metal : temporary workaround for the concurrency optimization bug

* falcon : add CUDA offloading (#2739)

* llama : better model naming and size reporting

* llama : prep new tokenizer support

* llama : advanced BPE tokenizer based on ggllm.cpp imlpementation

* llama : remove oboslete comment

ggml-ci

* common : remove obsolete BPE API + disable test-tokenizer-1

* llama : revert BPE special-case in llama_byte_to_token()

* cuda : add TODOs for RoPE NeoX implementation

* llama : default special tokens based on vocab type

* perplexity : add log for start of tokenization

---------

Co-authored-by: klosax <131523366+klosax@users.noreply.github.com>
Co-authored-by: slaren <slarengh@gmail.com>

1 files changed, 13 insertions, 13 deletions

diff --git a/gguf.py b/gguf.py
index 9421080b..5c37f0f0 100755
--- a/gguf.py
+++ b/gguf.py
@@ -30,12 +30,12 @@ KEY_GENERAL_SOURCE_HF_REPO       = "general.source.hugginface.repository"
 KEY_GENERAL_FILE_TYPE            = "general.file_type"
 
 # LLM
-KEY_LLM_CONTEXT_LENGTH        = "{arch}.context_length"
-KEY_LLM_EMBEDDING_LENGTH      = "{arch}.embedding_length"
-KEY_LLM_BLOCK_COUNT           = "{arch}.block_count"
-KEY_LLM_FEED_FORWARD_LENGTH   = "{arch}.feed_forward_length"
-KEY_LLM_USE_PARALLEL_RESIDUAL = "{arch}.use_parallel_residual"
-KEY_LLM_TENSOR_DATA_LAYOUT    = "{arch}.tensor_data_layout"
+KEY_CONTEXT_LENGTH        = "{arch}.context_length"
+KEY_EMBEDDING_LENGTH      = "{arch}.embedding_length"
+KEY_BLOCK_COUNT           = "{arch}.block_count"
+KEY_FEED_FORWARD_LENGTH   = "{arch}.feed_forward_length"
+KEY_USE_PARALLEL_RESIDUAL = "{arch}.use_parallel_residual"
+KEY_TENSOR_DATA_LAYOUT    = "{arch}.tensor_data_layout"
 
 # attention
 KEY_ATTENTION_HEAD_COUNT        = "{arch}.attention.head_count"
@@ -583,7 +583,7 @@ class GGUFWriter:
         self.add_string(KEY_GENERAL_AUTHOR, author)
 
     def add_tensor_data_layout(self, layout: str):
-        self.add_string(KEY_LLM_TENSOR_DATA_LAYOUT.format(arch=self.arch), layout)
+        self.add_string(KEY_TENSOR_DATA_LAYOUT.format(arch=self.arch), layout)
 
     def add_url(self, url: str):
         self.add_string(KEY_GENERAL_URL, url)
@@ -613,27 +613,27 @@ class GGUFWriter:
 
     def add_context_length(self, length: int):
         self.add_uint32(
-            KEY_LLM_CONTEXT_LENGTH.format(arch=self.arch), length)
+            KEY_CONTEXT_LENGTH.format(arch=self.arch), length)
 
     def add_embedding_length(self, length: int):
         self.add_uint32(
-            KEY_LLM_EMBEDDING_LENGTH.format(arch=self.arch), length)
+            KEY_EMBEDDING_LENGTH.format(arch=self.arch), length)
 
     def add_block_count(self, length: int):
         self.add_uint32(
-            KEY_LLM_BLOCK_COUNT.format(arch=self.arch), length)
+            KEY_BLOCK_COUNT.format(arch=self.arch), length)
 
     def add_feed_forward_length(self, length: int):
         self.add_uint32(
-            KEY_LLM_FEED_FORWARD_LENGTH.format(arch=self.arch), length)
+            KEY_FEED_FORWARD_LENGTH.format(arch=self.arch), length)
 
     def add_parallel_residual(self, use: bool):
         self.add_bool(
-            KEY_LLM_USE_PARALLEL_RESIDUAL.format(arch=self.arch), use)
+            KEY_USE_PARALLEL_RESIDUAL.format(arch=self.arch), use)
 
     def add_tensor_data_layout(self, layout: str):
         self.add_string(
-            KEY_LLM_TENSOR_DATA_LAYOUT.format(arch=self.arch), layout)
+            KEY_TENSOR_DATA_LAYOUT.format(arch=self.arch), layout)
 
     def add_head_count(self, count: int):
         self.add_uint32(