quickjs bare-bones?

separate jamfile project, WIP

Makefile

@@ -16,6 +16,7 @@ include llamafile/BUILD.mk
 include llama.cpp/BUILD.mk
 include stable-diffusion.cpp/BUILD.mk
 include whisper.cpp/BUILD.mk
+include jamfile/BUILD.mk
 
 # the root package is `o//` by default
 # building a package also builds its sub-packages

jamfile/.clang-format

@@ -0,0 +1,12 @@
+---
+BasedOnStyle: LLVM
+IndentWidth: 4
+ColumnLimit: 100
+---
+Language: Cpp
+AllowShortFunctionsOnASingleLine: false
+AlignTrailingComments: false
+AlignEscapedNewlines: DontAlign
+AlwaysBreakTemplateDeclarations: true
+ConstructorInitializerAllOnOneLineOrOnePerLine: true
+---

jamfile/BUILD.mk

@@ -0,0 +1,110 @@
+#-*-mode:makefile-gmake;indent-tabs-mode:t;tab-width:8;coding:utf-8-*-┐
+#── vi: set noet ft=make ts=8 sw=8 fenc=utf-8 :vi ────────────────────┘
+
+PKGS += LLAMA_CPP_JAMFILE
+
+LLAMA_CPP_JAMFILE_FILES := $(wildcard jamfile/*)
+LLAMA_CPP_JAMFILE_HDRS = $(filter %.h,$(LLAMA_CPP_JAMFILE_FILES))
+LLAMA_CPP_JAMFILE_SRCS_C = $(filter %.c,$(LLAMA_CPP_JAMFILE_FILES))
+LLAMA_CPP_JAMFILE_SRCS_CPP = $(filter %.cpp,$(LLAMA_CPP_JAMFILE_FILES))
+LLAMA_CPP_JAMFILE_SRCS = $(LLAMA_CPP_JAMFILE_SRCS_C) $(LLAMA_CPP_JAMFILE_SRCS_CPP)
+
+LLAMA_CPP_JAMFILE_OBJS = \
+	$(LLAMA_CPP_JAMFILE_SRCS_C:%.c=o/$(MODE)/%.o) \
+	$(LLAMA_CPP_JAMFILE_SRCS_CPP:%.cpp=o/$(MODE)/%.o)
+
+
+#o/$(MODE)/jamfile/jamfile.a: $(LLAMA_CPP_JAMFILE_SRCS_C)
+
+### o/$(MODE)/jamfile/sqlite-vec.o: jamfile/sqlite-vec.c
+### o/$(MODE)/jamfile/sqlite-vec.a: o/$(MODE)/jamfile/sqlite-vec.o
+### 
+### o/$(MODE)/jamfile/sqlite-csv.o: jamfile/sqlite-csv.c
+### o/$(MODE)/jamfile/sqlite-csv.a: o/$(MODE)/jamfile/sqlite-csv.o
+### 
+### 
+o/$(MODE)/jamfile/quickjs-llamafile-completion.o: jamfile/quickjs-llamafile-completion.cpp
+o/$(MODE)/jamfile/quickjs-llamafile-completion.a: o/$(MODE)/jamfile/quickjs-llamafile-completion.o
+
+o/$(MODE)/jamfile/quickjs-llamafile.o: jamfile/quickjs-llamafile.c jamfile/quickjs-llamafile-completion.cpp #o/$(MODE)/jamfile/quickjs-llamafile-completion.o
+o/$(MODE)/jamfile/quickjs-llamafile.a: o/$(MODE)/jamfile/quickjs-llamafile.o o/$(MODE)/jamfile/quickjs-llamafile-completion.a
+### 
+### o/$(MODE)/jamfile/sqlite-lines.o: jamfile/sqlite-lines.c
+### o/$(MODE)/jamfile/sqlite-lines.a: o/$(MODE)/jamfile/sqlite-lines.o
+### 
+### o/$(MODE)/jamfile/sqlite-lembed.o: jamfile/sqlite-lembed.c
+### o/$(MODE)/jamfile/sqlite-lembed.a: o/$(MODE)/jamfile/sqlite-lembed.o o/$(MODE)/llama.cpp/llama.cpp.a
+
+
+# jamfile:assert bytecode
+o/$(MODE)/jamfile/assert.gen.c: jamfile/js_builtins/assert.js o/$(mode)/third_party/quickjs/qjsc
+	o/$(mode)/third_party/quickjs/qjsc -m -o $@ $<
+o/$(MODE)/jamfile/assert.gen.o: o/$(MODE)/jamfile/assert.gen.c
+
+# jamfile:fmt bytecode
+o/$(MODE)/jamfile/fmt.gen.c: jamfile/js_builtins/fmt.js o/$(mode)/third_party/quickjs/qjsc
+	o/$(mode)/third_party/quickjs/qjsc -m -o $@ $<
+o/$(MODE)/jamfile/fmt.gen.o: o/$(MODE)/jamfile/fmt.gen.c
+
+# jamfile:cli bytecode
+o/$(MODE)/jamfile/cli.gen.c: jamfile/js_builtins/cli.js o/$(MODE)/third_party/quickjs/qjsc
+	o/$(mode)/third_party/quickjs/qjsc -m -o $@ $<
+o/$(MODE)/jamfile/cli.gen.o: o/$(MODE)/jamfile/cli.gen.c
+
+# jamfile:colors bytecode
+o/$(MODE)/jamfile/colors.gen.c: jamfile/js_builtins/colors.js o/$(MODE)/third_party/quickjs/qjsc
+	o/$(mode)/third_party/quickjs/qjsc -m -o $@ $<
+o/$(MODE)/jamfile/colors.gen.o: o/$(MODE)/jamfile/colors.gen.c
+
+# jamfile:zod bytecode
+o/$(MODE)/jamfile/zod.gen.c: jamfile/js_builtins/zod.js o/$(mode)/third_party/quickjs/qjsc
+	o/$(mode)/third_party/quickjs/qjsc -m -o $@ $<
+o/$(MODE)/jamfile/zod.gen.o: o/$(MODE)/jamfile/zod.gen.c
+
+# include the raw contents of the jamfile.d.ts into jamfile for the `jamfile types` command
+o/$(MODE)/jamfile/jamfile-types.c: jamfile/jamfile.d.ts
+	xxd --include $< > $@
+o/$(MODE)/jamfile/jamfile-types.o: o/$(MODE)/jamfile/jamfile-types.c
+
+o/$(MODE)/jamfile/jamfile.a: \
+	o/$(MODE)/jamfile/jamfile-types.o \
+	o/$(MODE)/jamfile/cli.gen.o		\
+	o/$(MODE)/jamfile/fmt.gen.o		\
+	o/$(MODE)/jamfile/zod.gen.o		\
+	o/$(MODE)/jamfile/assert.gen.o		\
+	o/$(MODE)/jamfile/colors.gen.o		\
+	o/$(MODE)/jamfile/quickjs-sqlite.o		\
+	o/$(MODE)/jamfile/quickjs-llamafile.o		\
+	o/$(MODE)/jamfile/quickjs-llamafile-completion.o		\
+	o/$(MODE)/third_party/quickjs/repl.o		\
+	o/$(MODE)/third_party/quickjs/cutils.o	\
+	o/$(MODE)/third_party/quickjs/libbf.o	\
+	o/$(MODE)/third_party/quickjs/quickjs.o		\
+	o/$(MODE)/third_party/quickjs/libregexp.o		\
+	o/$(MODE)/third_party/quickjs/libunicode.o		\
+	o/$(MODE)/third_party/quickjs/quickjs-libc.o
+
+
+o/$(MODE)/jamfile/jamfile:					\
+		o/$(MODE)/jamfile/jamfile.a \
+		o/$(MODE)/jamfile/jamfile.1.asc.zip.o	\
+		o/$(MODE)/llama.cpp/llama.cpp.a \
+		o/$(MODE)/third_party/sqlite/sqlite3.a \
+		o/$(MODE)/jamfile/quickjs-sqlite.a \
+		o/$(MODE)/jamfile/quickjs-llamafile.a \
+		o/$(MODE)/embedfile/sqlite-csv.a \
+		o/$(MODE)/embedfile/sqlite-vec.a \
+		o/$(MODE)/embedfile/sqlite-lines.a \
+		o/$(MODE)/embedfile/sqlite-lembed.a
+
+$(LLAMA_CPP_JAMFILE_OBJS): private CCFLAGS += -DSQLITE_CORE
+
+.PHONY: o/$(MODE)/jamfile
+o/$(MODE)/jamfile:						\
+		o/$(MODE)/jamfile/jamfile
+
+$(LLAMA_CPP_JAMFILE_OBJS): llama.cpp/BUILD.mk jamfile/BUILD.mk
+.PHONY: jamfile-test-js
+
+jamfile-test-js: o/$(MODE)/jamfile/jamfile jamfile/tests/js/test.js
+	$< run jamfile/tests/js/test.js

jamfile/examples/spellcheck/document.txt

@@ -0,0 +1,6 @@
+The first spell ceckers were "verifiers" instead of "correctors." They
+offered no suggestions for incorrectly spelled words. This was helpful
+for typos but it was not so helpfl for logical or phonetic errors. The
+challenge the developers faced was the difficulty in offering useful
+suggestions for misspelled words. This requires reducing words to a
+sceletal form and applying pattern-matching algorithms.

jamfile/examples/spellcheck/spellcheck.gbnf

@@ -0,0 +1,4 @@
+root ::= good | list
+good ::= "ok"
+list ::= item ( "\n" item )*
+item ::= "- `" [-_0-9A-Za-z]+ "` should be `" [-_0-9A-Za-z]+ "`"

jamfile/examples/spellcheck/spellcheck.js

@@ -0,0 +1,36 @@
+/// <reference types="./jamfile/jamfile.d.ts"/>
+
+import { CompletionModel } from "jamfile:llamafile";
+import schema from "./spellcheck.gbnf";
+
+const model = new CompletionModel('gemma-2-2b-it.Q2_K.llamafile');
+
+let prompt = (document) => `
+<start_of_turn>user
+You are a helpful AI assistant that does spell checking and only spell
+checking. For example, if the user says:
+
+It also meas that I keep my opshuns open.
+
+Then the assistant will say:
+
+- \`meas\` should be \`means\`
+- \`opshuns\` should be \`options\`
+
+If there are no mistyped words, the assistant responds with:
+
+ok
+
+Never reply with ok.<end_of_turn>
+<start_of_turn>user
+${document}<end_of_turn>
+<start_of_turn>assistant
+`;
+
+
+import * as std from "qjs:std";
+
+const document = std.in.readAsString();
+const response = model.complete(prompt(document), { schema });
+
+console.log(response);

jamfile/examples/sqlite-vec/demo.js

@@ -0,0 +1,60 @@
+import {open} from "qjs:std";
+import {TextEmbeddingModel} from "jamfile:llamafile";
+import {Database} from "jamfile:sqlite";
+import {formatDuration} from "jamfile:fmt";
+import {green} from "jamfile:colors";
+
+const README = open('README.md', 'r').readAsString();
+
+const db = new Database();
+const model = new TextEmbeddingModel('dist/.models/mxbai-embed-xsmall-v1-f16.gguf');
+
+db.execute(`
+  CREATE VIRTUAL TABLE vec_chunks USING vec0(
+    +contents TEXT,
+    contents_embedding float[384] distance_metric=cosine
+  );`);
+
+
+function* chunks(arr, n) {
+  for (let i = 0; i < arr.length; i += n) {
+    yield arr.slice(i, i + n);
+  }
+}
+
+const tokens = model.tokenize(README);
+
+const t0 = Date.now();
+const c = Array.from(chunks(tokens, 64));
+for(const chunk of c) {
+  const chunk_embedding = model.embed(chunk);
+  db.execute(
+    'INSERT INTO vec_chunks(contents, contents_embedding) VALUES (?, ?);',
+    [model.detokenize(chunk), chunk_embedding]
+  );
+}
+const duration = Date.now() - t0;
+
+console.log(`${formatDuration(duration)} for ${c.length} embeddings (${c.length / (duration / 1000)}/second)`)
+
+function search(query) {
+  const rows = db.queryAll(
+    `SELECT
+      rowid, 
+      contents 
+    FROM vec_chunks 
+    WHERE contents_embedding MATCH ? 
+      AND k = 10
+    `, 
+    [model.embed(query)]
+  );
+
+  console.log(green(query));
+
+  for (const {rowid, contents} of rows) {
+    console.log(contents);
+  }
+}
+
+search('linux CLI Tools ');
+search('money supporting ecosystem ');

jamfile/examples/standlone/.gitignore

@@ -0,0 +1 @@
+*.jamfile

jamfile/examples/standlone/Makefile

@@ -0,0 +1,25 @@
+Llama-3.2-1B-Instruct-Q4_K_M.gguf:
+	curl -L -o $@ https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/$@
+
+mxbai-embed-xsmall-v1-f16.gguf:
+	curl -L -o $@ https://huggingface.co/mixedbread-ai/mxbai-embed-xsmall-v1/resolve/main/gguf/$@
+
+standalone.jamfile:  \
+	../../../o/jamfile/jamfile \
+	Llama-3.2-1B-Instruct-Q4_K_M.gguf \
+	mxbai-embed-xsmall-v1-f16.gguf \
+	standalone.js 
+
+	printf "%s\n" \
+		"--default-embedding-model" \
+		"mxbai-embed-xsmall-v1-f16.gguf" \
+		"--default-completion-model" \
+		"Llama-3.2-1B-Instruct-Q4_K_M.gguf" \
+		"run" \
+		"standalone.js" \
+		"..." > .args
+
+	cp ../../../o/jamfile/jamfile $@
+	../../../o/llamafile/zipalign -j0 $@ $^ .args
+	rm .args
+	chmod u+x $@

jamfile/examples/standlone/README.md

@@ -0,0 +1,36 @@
+# jamfile sample: standalone
+
+An example of how you can make a standlone single-binary script from a Jamfile JavaScript script.
+
+The `standalone.js` is a sample JavaScript script that generates a a sample embedding and completion result. It could be ran as-is with `jamfile run standlone.js`, but we could bundle the JS script and model weights into a single-file `jamfile` that works on all operating systems.
+
+To do so, run the following command:
+
+```
+make sample.jamfile
+```
+
+That will download an embeddings model (MixedBread xsmall) and a LLM (Lllama 3.2), embedding them into a new `sample.jamfile` file with the `standalone.js` JavaScript script.
+
+You can now execute that `jamfile` directly like so:
+
+
+```
+$ ./standalone.jamfile 
+embedding sample:  [-0.05826485529541969,0.04374322667717934,0.03084852732717991,0.047234565019607544,-0.05963338911533356,-0.03614785894751549,0.03814202547073364,0.005149350967258215]
+completion sample:  
+Here is a single haiku about SpongeBob SquarePants:
+
+Fluffy sponge so bright
+Rainbow colors in his heart
+Joyful underwater
+
+I hope you like it! Let me know if you have any other requests.
+```
+
+The output `standalone.jamfile` is a small `850MB`, which includes the LLM, embeddings model, and `jamfile` executable all in one.
+
+- `jamfile` base: `15MB`
+- `mxbai-embed-xsmall-v1-f16.gguf`: `49MB`
+- `Llama-3.2-1B-Instruct-Q4_K_M.gguf`: `770MB`
+- `standalone.jamfile`: `850MB`

jamfile/examples/standlone/standalone.js

@@ -0,0 +1,23 @@
+/**
+ * An example of a standalone jamfile. 
+ * 
+ * This script performs 
+ */
+
+/// <reference types="../../jamfile.d.ts"/>
+
+import { red } from "jamfile:color";
+import { CompletionModel, TextEmbeddingModel } from "jamfile:llamafile";
+
+const embeddingModel = new TextEmbeddingModel();
+const completionModel = new CompletionModel();
+
+console.log(
+  red("embedding sample: "), 
+  JSON.stringify(Array.from(embeddingModel.embed("hello!").slice(0, 8)))
+);
+
+console.log(
+  red("completion sample: "), 
+  completionModel.complete("write a single haiku about spongebob squarepants")
+);