Merge pull request 'Refresh sources' (#2) from refresh_sources into main
Reviewed-on: #2
This commit was merged in pull request #2.
This commit is contained in:
+2
-2
@@ -1,3 +1,3 @@
|
||||
.DS_Store
|
||||
PRIVATE_DATA/**
|
||||
!PRIVATE_DATA/.gitkeep
|
||||
PRIVATE_DATA/
|
||||
config.json
|
||||
|
||||
@@ -4,20 +4,60 @@ Finding useful things to do with local LLMs.
|
||||
All scripts work with data within `PRIVATE_DATA/` so that private data can't be
|
||||
accidentally committed.
|
||||
|
||||
###### `cosine_similarity.lua`
|
||||
### `cosine_similarity.lua`
|
||||
Opens `embeddings.json` and creates `similarities.json` with a sorted list of
|
||||
comparisons.
|
||||
|
||||
###### `generate_embeddings.lua`
|
||||
### `generate_embeddings.lua`
|
||||
Opens every file on its whitelist within the `notebook` directory, and generates
|
||||
embeddings (placed in `embeddings.json`). When files are too long, it truncates
|
||||
them.
|
||||
|
||||
###### `least_similar.lua`
|
||||
### `least_similar.lua`
|
||||
Opens `similarities.json`, reverses the sort order, and saves it as
|
||||
`differences.json`.
|
||||
|
||||
###### `synopsis_generator.lua`
|
||||
### `refresh_sources.lua`
|
||||
Creates/Maintains a store of chunked data with embeddings based on configurable
|
||||
sources. Example:
|
||||
|
||||
```json
|
||||
{
|
||||
"source name":{
|
||||
"filters":{
|
||||
"blacklist":[".git"],
|
||||
"extension_whitelist":["md"]
|
||||
},
|
||||
"initialize_command":"git clone REMOTE .",
|
||||
"path":"will be created before initialize_command is run",
|
||||
"strip_frontmatter":true,
|
||||
"refresh_command":"git reset --hard origin/main"
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
The filters are based on `utility.tree`'s filter options (optional).
|
||||
`initialize_command` is only run the first time (optional),
|
||||
while `refresh_command` is run each time (optional).
|
||||
Commands will be run in the specified `path`.
|
||||
`strip_frontmatter` will remove YAML frontmatter (common in Markdown files).
|
||||
|
||||
The embeddings are stored like so:
|
||||
|
||||
```json
|
||||
{
|
||||
"files":{
|
||||
"PRIVATE_DATA/source_path/path/to/file.ext":["sha512sum", "another sum"]
|
||||
},
|
||||
"vectors":{
|
||||
"sha512sum":[0.5, 0, 1, -0.5, -1, ...]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
TODO
|
||||
|
||||
### `synopsis_generator.lua`
|
||||
Chooses a random file within `notebook`, and generates a novel synopsis from it.
|
||||
|
||||
Arguments:
|
||||
|
||||
+75
@@ -0,0 +1,75 @@
|
||||
local stored_messages = {}
|
||||
local display_levels = {}
|
||||
local enable_logging = true
|
||||
local message_limit = math.huge
|
||||
|
||||
-- USAGE:
|
||||
-- -- set arbitrary log levels to be printed immediately
|
||||
-- log{ info = true, warning = true, ducksauce = true, }
|
||||
-- -- send anything (except nil) to a log level
|
||||
-- log("bacom", true, "text", 5, function() end, {})
|
||||
-- -- print all messages saved at a particular log level
|
||||
-- log("error")
|
||||
-- -- toggle ALL logging on and off with a boolean
|
||||
-- log(false) -- this deletes all stored messages
|
||||
-- -- set a maximum number of stored messages per log level
|
||||
-- log(50) -- warning: reaching large limits makes logging expensive
|
||||
-- -- do anything to the stored messages (this example deletes them all)
|
||||
-- log(function(messages) return {} end)
|
||||
|
||||
return function(options, ...)
|
||||
local options_type = type(options)
|
||||
|
||||
if (options_type == "string") and enable_logging then
|
||||
if not stored_messages[options] then stored_messages[options] = {} end
|
||||
local message_table = stored_messages[options]
|
||||
|
||||
-- turn log message into text
|
||||
local tab = {...}
|
||||
for i = 1, #tab do
|
||||
tab[i] = tostring(tab[i])
|
||||
end
|
||||
local current_message = table.concat(tab, "\t")
|
||||
|
||||
if #current_message == 0 then
|
||||
-- print all stored_messages of specified level
|
||||
for i = 1, #message_table do
|
||||
print(message_table[i])
|
||||
end
|
||||
|
||||
else
|
||||
-- store message (and print if in display_levels)
|
||||
message_table[#message_table + 1] = current_message
|
||||
if #message_table > message_limit then table.remove(message_table, 1) end
|
||||
if display_levels[options] then print(current_message) end
|
||||
end
|
||||
|
||||
-- set display_levels
|
||||
elseif options_type == "table" then
|
||||
for k,v in pairs(options) do
|
||||
display_levels[k] = v
|
||||
end
|
||||
|
||||
elseif options_type == "number" then
|
||||
message_limit = options
|
||||
for k,v in pairs(stored_messages) do
|
||||
while #v > message_limit do
|
||||
table.remove(v, 1)
|
||||
end
|
||||
end
|
||||
|
||||
-- turn all logging off and delete stored messages; or turn it on
|
||||
elseif options_type == "boolean" then
|
||||
if options then
|
||||
enable_logging = true
|
||||
else
|
||||
enable_logging = false
|
||||
stored_messages = {}
|
||||
end
|
||||
|
||||
-- arbitrary access :D
|
||||
elseif options_type == "function" then
|
||||
local result = options(stored_messages)
|
||||
if type(result) == "table" then stored_messages = result end
|
||||
end
|
||||
end
|
||||
+23
-8
@@ -1,18 +1,26 @@
|
||||
local timing = {}
|
||||
|
||||
local function human_readable_time(delta)
|
||||
if delta >= 2*7*24*60*60 then -- if more than 2 weeks
|
||||
delta = tostring(math.floor(delta/(7*24*60*60))/10) .. " weeks"
|
||||
elseif delta >= 2*24*60*60 then -- if more than 2 days
|
||||
delta = tostring(math.floor(delta/(24*60*60))/10) .. " days"
|
||||
elseif delta >= 2*60*60 then -- if more than 2 hours
|
||||
delta = tostring(math.floor(delta/(60*60/10))/10) .. " hours"
|
||||
elseif delta >= 2*60 then -- if more then 2 minutes
|
||||
delta = tostring(math.floor(delta/(60/10))/10) .. " minutes"
|
||||
else
|
||||
delta = tostring(delta) .. " seconds"
|
||||
end
|
||||
return delta
|
||||
end
|
||||
|
||||
timing.display = function(n)
|
||||
local function _display(n)
|
||||
local time = timing[n]
|
||||
local previous = timing[n - 1]
|
||||
|
||||
local delta = time.time - previous.time
|
||||
if delta >= 2*60*60 then -- if more than 2 hours
|
||||
delta = tostring(math.floor(delta/(60*60/10))/10) .. " hours"
|
||||
elseif delta >= 120 then -- if more then 2 minutes
|
||||
delta = tostring(math.floor(delta/(60/10))/10) .. " minutes"
|
||||
else
|
||||
delta = tostring(delta) .. " seconds"
|
||||
end
|
||||
local delta = human_readable_time(time.time - previous.time)
|
||||
|
||||
print(delta, previous.label)
|
||||
end
|
||||
@@ -32,6 +40,13 @@ timing.mark = function(label)
|
||||
if #timing > 1 then
|
||||
timing.display(#timing)
|
||||
end
|
||||
print("", "", label)
|
||||
end
|
||||
|
||||
timing.estimate = function(current_position, total_operations)
|
||||
local delta = os.time() - timing[#timing].time
|
||||
local estimate = delta * total_operations / current_position - delta
|
||||
return human_readable_time(estimate)
|
||||
end
|
||||
|
||||
return timing
|
||||
|
||||
+41
-12
@@ -18,7 +18,7 @@ if package.config:sub(1, 1) == "\\" then
|
||||
}
|
||||
else
|
||||
utility = {
|
||||
OS = "UNIX-like",
|
||||
OS = "Linux",
|
||||
path_separator = "/",
|
||||
temp_directory = "/tmp/",
|
||||
commands = {
|
||||
@@ -81,6 +81,10 @@ standard_library_addition(string, "split", function(s, delimiter)
|
||||
return result
|
||||
end)
|
||||
|
||||
utility.leftpad = function(text, length, character)
|
||||
return string.rep(character or " ", length - #(tostring(text))) .. text
|
||||
end
|
||||
|
||||
|
||||
|
||||
utility.require = function(...)
|
||||
@@ -217,7 +221,7 @@ utility.list = function(path, func)
|
||||
|
||||
local run = function(fn)
|
||||
for line in output:gmatch("[^\r\n]+") do -- thanks to https://stackoverflow.com/a/32847589
|
||||
if not (line == "." or line == "..") then
|
||||
if not ((line == ".") or (line == "..")) then
|
||||
fn(line)
|
||||
end
|
||||
end
|
||||
@@ -235,7 +239,8 @@ utility.ls = function(...)
|
||||
return utility.list(...)
|
||||
end
|
||||
|
||||
utility.tree = function(path, options, fn)
|
||||
local tree
|
||||
tree = function(path, options, fn)
|
||||
if type(options) == "function" then
|
||||
fn = options
|
||||
options = {}
|
||||
@@ -245,16 +250,25 @@ utility.tree = function(path, options, fn)
|
||||
if options.blacklist and options.blacklist[path_name] then return end
|
||||
if options.whitelist and (not options.whitelist[path_name]) then return end
|
||||
|
||||
if options.extension_blacklist or options.extension_whitelist then
|
||||
local _, _, extension = utility.split_path_components(path_name)
|
||||
if options.extension_blacklist and options.extension_blacklist[extension] then return end
|
||||
if options.extension_whitelist and (not options.extension_whitelist[extension]) then return end
|
||||
end
|
||||
|
||||
if utility.is_file(path_name) then
|
||||
if options.extension_blacklist or options.extension_whitelist then
|
||||
local _, _, extension = utility.split_path_components(path_name)
|
||||
if options.extension_blacklist and options.extension_blacklist[extension] then return end
|
||||
if options.extension_whitelist and (not options.extension_whitelist[extension]) then return end
|
||||
end
|
||||
|
||||
fn(path_name)
|
||||
else
|
||||
utility.tree(path .. utility.path_separator .. path_name, options, fn)
|
||||
tree(path .. utility.path_separator .. path_name, options, fn)
|
||||
end
|
||||
end)
|
||||
end
|
||||
utility.tree = function(path, options, fn)
|
||||
tree(path, options, function(path_name)
|
||||
if path_name:find(path) == 1 then
|
||||
fn(path_name)
|
||||
else
|
||||
fn(path .. utility.path_separator .. path_name)
|
||||
end
|
||||
end)
|
||||
end
|
||||
@@ -265,10 +279,11 @@ utility.read_file = function(file_name)
|
||||
end)
|
||||
end
|
||||
|
||||
utility.write_file = function(file_name, text)
|
||||
utility.write_file = function(file_name, ...)
|
||||
local text = table.concat{...}
|
||||
return utility.open(file_name, "w", function(file)
|
||||
file:write(text)
|
||||
file:write("\n")
|
||||
-- file:write("\n") -- I need to make sure /I/ handle this instead of trying to automate it
|
||||
end)
|
||||
end
|
||||
|
||||
@@ -299,6 +314,16 @@ utility.file_size = function(file_path)
|
||||
return utility.open(file_path, "rb", function(file) return file:seek("end") end)
|
||||
end
|
||||
|
||||
utility.sha512sum = function(file_path)
|
||||
local sha512sum
|
||||
if (utility.OS == "Linux") or (utility.OS == "macOS") then
|
||||
sha512sum = utility.capture_safe("shasum -U -a 512 " .. file_path:enquote())
|
||||
elseif utility.OS == "Windows" then
|
||||
error("utility.sha512sum() not implemented for Windows.")
|
||||
end
|
||||
return sha512sum:sub(1, 128)
|
||||
end
|
||||
|
||||
|
||||
|
||||
utility.escape_quotes_and_escapes = function(input)
|
||||
@@ -570,4 +595,8 @@ end
|
||||
|
||||
|
||||
|
||||
if (utility.OS == "Linux") and (utility.capture_safe("uname"):find("Darwin") == 1) then
|
||||
utility.OS = "macOS"
|
||||
end
|
||||
|
||||
return utility
|
||||
|
||||
Executable
+262
@@ -0,0 +1,262 @@
|
||||
#!/usr/bin/env luajit
|
||||
|
||||
package.path = (arg[0]:match("@?(.*/)") or arg[0]:match("@?(.*\\)")) .. "lib" .. package.config:sub(1, 1) .. "?.lua;" .. package.path
|
||||
local utility = require "utility"
|
||||
|
||||
local json = utility.require("dkjson")
|
||||
local log = utility.require("log")
|
||||
local text_processing = utility.require("text_processing")
|
||||
local timing = utility.require("timing")
|
||||
|
||||
-- TODO make utility have a function for getting/setting defaults where locking is only used to set defaults if they aren't present
|
||||
-- TODO there should be a check function for a lock so a warning/error can be dumped?
|
||||
local config = utility.get_config("no-lock")
|
||||
if not config.models then
|
||||
config.models = {
|
||||
embedding = {
|
||||
model = "qwen3-embedding:0.6b",
|
||||
max_chunk_size = 32768,
|
||||
},
|
||||
initialized_sources = {},
|
||||
}
|
||||
utility.save_config()
|
||||
end
|
||||
|
||||
log{
|
||||
info = true,
|
||||
warning = true,
|
||||
-- debug = true,
|
||||
-- files = true, -- debugging why the wrong files are selected
|
||||
-- sha = true, -- what the fuck is going on with sha sums?
|
||||
}
|
||||
|
||||
local memory_path = "PRIVATE_DATA" .. utility.path_separator .. "memory"
|
||||
local embeddings_file_path = memory_path .. utility.path_separator .. "+embeddings.json"
|
||||
local tmp_file_path = "PRIVATE_DATA" .. utility.path_separator .. ".tmp.2b65c19b-0883-49ca-8247-b1fe7760f922"
|
||||
|
||||
local embeddings
|
||||
if not utility.path_exists(embeddings_file_path) then
|
||||
os.execute("mkdir -p " .. memory_path:enquote())
|
||||
utility.save_data({
|
||||
files = {},
|
||||
vectors = {},
|
||||
}, embeddings_file_path)
|
||||
end
|
||||
embeddings = utility.load_data(embeddings_file_path)
|
||||
local function embeddings_debug()
|
||||
log("debug", "Embeddings loaded.", embeddings, embeddings.files, embeddings.vectors)
|
||||
local file_count = 0
|
||||
for k,v in pairs(embeddings.files) do
|
||||
file_count = file_count + 1
|
||||
end
|
||||
local vector_count = 0
|
||||
for k,v in pairs(embeddings.vectors) do
|
||||
vector_count = vector_count + 1
|
||||
end
|
||||
log("debug", file_count .. " files.")
|
||||
log("debug", vector_count .. " vectors.")
|
||||
-- os.exit(1)
|
||||
end
|
||||
embeddings_debug()
|
||||
|
||||
|
||||
|
||||
local refresh_file_list = function(source_name, data_source)
|
||||
timing.mark("Assembling file list for " .. source_name .. ".")
|
||||
|
||||
local full_path = "PRIVATE_DATA" .. utility.path_separator .. data_source.path
|
||||
if data_source.initialize_command and (not config.models.initialized_sources[data_source.path]) then
|
||||
-- NOTE this is where we'd want to check/obtain a lock on the config so we can safely run this
|
||||
os.execute("mkdir -p " .. full_path:enquote() .. " && cd " .. full_path:enquote() .. " && " .. data_source.initialize_command)
|
||||
config.models.initialized_sources[data_source.path] = true
|
||||
utility.save_config()
|
||||
end
|
||||
if data_source.refresh_command then
|
||||
os.execute("cd " .. full_path:enquote() .. " && " .. data_source.refresh_command)
|
||||
end
|
||||
|
||||
local compiled_filters = {}
|
||||
if data_source.filters then
|
||||
for name, object in pairs(data_source.filters) do
|
||||
compiled_filters[name] = utility.enumerate(object)
|
||||
end
|
||||
end
|
||||
|
||||
local file_list = {}
|
||||
utility.tree("PRIVATE_DATA" .. utility.path_separator .. data_source.path, compiled_filters, function(file_name)
|
||||
log("files", file_name)
|
||||
file_list[#file_list + 1] = file_name
|
||||
end)
|
||||
|
||||
timing.mark("Finished assembling file list for " .. source_name .. ".")
|
||||
return file_list
|
||||
end
|
||||
|
||||
-- returns nothing when too much text is sent
|
||||
local generate_embeddings = function(text)
|
||||
if #text > config.models.embedding.max_chunk_size then
|
||||
return nil, "generate_embeddings() must only be passed appropriately-sized chunks!"
|
||||
end
|
||||
|
||||
local result = utility.llm_prompt(text, config.models.embedding.model)
|
||||
return json.decode(result)
|
||||
end
|
||||
|
||||
-- returns nothing for empty files and errors
|
||||
local process_file = function(data_source, file_name)
|
||||
local text = utility.read_file(file_name)
|
||||
|
||||
if data_source.strip_frontmatter then
|
||||
text = text_processing.strip_frontmatter(text)
|
||||
end
|
||||
|
||||
if #text == 0 then
|
||||
log("empty", file_name .. "\n is empty and being skipped.")
|
||||
return
|
||||
end
|
||||
|
||||
local chunk_size = config.models.embedding.max_chunk_size
|
||||
local half_chunk_size = math.floor(chunk_size / 2)
|
||||
local chunks = { text }
|
||||
|
||||
while #text > chunk_size do
|
||||
local first_chunk = text:sub(1, chunk_size)
|
||||
local overlap_chunk = text:sub(half_chunk_size, chunk_size + half_chunk_size - 1)
|
||||
|
||||
chunks[#chunks + 1] = first_chunk
|
||||
chunks[#chunks + 1] = overlap_chunk
|
||||
|
||||
text = text:sub(chunk_size)
|
||||
if (#text > half_chunk_size) and (not (#text > chunk_size)) then
|
||||
-- last chunk would be skipped if we didn't handle this here
|
||||
chunks[#chunks + 1] = text
|
||||
end
|
||||
end
|
||||
|
||||
local new_embeddings = {}
|
||||
for i = 1, #chunks do
|
||||
log("debug", "Embedding length:", #chunks[i])
|
||||
new_embeddings[i] = generate_embeddings(chunks[i]) or {}
|
||||
end
|
||||
|
||||
if #new_embeddings[1] == 0 then
|
||||
log("debug", "Vector lengths:")
|
||||
for e = 1, #new_embeddings do
|
||||
log("debug", "", e, #new_embeddings[e])
|
||||
end
|
||||
if #new_embeddings == 1 then
|
||||
-- Ollama very rarely errors with:
|
||||
-- Error: do embedding request: Post "http://127.0.0.1:53441/v1/embeddings": EOF
|
||||
-- but it is inconsistent and re-running will eventually fix it.
|
||||
log("warning", file_name .. "\n encountered an embedding error and will be skipped this run only.")
|
||||
return
|
||||
end
|
||||
|
||||
-- average all embeddings to make the core file embedding
|
||||
local count = #new_embeddings[2]
|
||||
for vector_index = 1, count do
|
||||
local total = 0
|
||||
for chunk = 2, #new_embeddings do
|
||||
total = total + new_embeddings[chunk][vector_index]
|
||||
end
|
||||
new_embeddings[1][vector_index] = total / count
|
||||
end
|
||||
end
|
||||
|
||||
return chunks, new_embeddings
|
||||
end
|
||||
|
||||
-- returns nothing for errors
|
||||
local memorize_file = function(data_source, file_name)
|
||||
local file_chunks, file_embeddings = process_file(data_source, file_name)
|
||||
if not file_chunks then return end
|
||||
|
||||
local file_sums = {}
|
||||
for i = 1, #file_chunks do
|
||||
local function loop()
|
||||
local text = file_chunks[i]
|
||||
local current_embedding = file_embeddings[i]
|
||||
|
||||
utility.write_file(tmp_file_path, text)
|
||||
|
||||
local sha512sum = utility.sha512sum(tmp_file_path)
|
||||
log("sha", "New sum? " .. sha512sum, tostring(embeddings.vectors[sha512sum]))
|
||||
file_sums[#file_sums + 1] = sha512sum
|
||||
if embeddings.vectors[sha512sum] then
|
||||
log("sha", "Detected previously extant sum, skipping.")
|
||||
return
|
||||
end
|
||||
|
||||
os.execute(utility.commands.move .. tmp_file_path:enquote()
|
||||
.. " " .. (memory_path .. utility.path_separator .. sha512sum):enquote())
|
||||
embeddings.vectors[sha512sum] = current_embedding
|
||||
log("sha", "Saved new sum!")
|
||||
end
|
||||
loop()
|
||||
end
|
||||
|
||||
return file_sums
|
||||
end
|
||||
|
||||
local refresh_sources = function()
|
||||
os.execute("mkdir -p " .. memory_path:enquote())
|
||||
|
||||
timing.mark("Checking memory for missing embeddings.")
|
||||
utility.list(memory_path, function(file_name)
|
||||
log("files", file_name)
|
||||
if (file_name == ".DS_Store") or (file_name == "+embeddings.json") then return end
|
||||
|
||||
local file_path = memory_path .. utility.path_separator .. file_name
|
||||
local sha512sum = utility.sha512sum(file_path)
|
||||
log("sha", "New sum? " .. sha512sum, tostring(embeddings.vectors[sha512sum]))
|
||||
if not embeddings.vectors[sha512sum] then
|
||||
log("sha", "Saved new sum!")
|
||||
memorize_file({}, file_path)
|
||||
end
|
||||
end)
|
||||
timing.mark("Finished checking memory for missing embeddings.")
|
||||
|
||||
local sources = utility.load_data("PRIVATE_DATA/sources.json")
|
||||
for source_name, data_source in pairs(sources) do
|
||||
local file_list = refresh_file_list(source_name, data_source)
|
||||
|
||||
timing.mark("Generating embeddings from source \"" .. source_name .. "\"")
|
||||
for f = 1, #file_list do
|
||||
local file_name = file_list[f]
|
||||
local function loop()
|
||||
local sha512sum = utility.sha512sum(file_name)
|
||||
log("sha", file_name, sha512sum, "\n Sum present? " .. tostring(embeddings.vectors[sha512sum]))
|
||||
if embeddings.vectors[sha512sum] then
|
||||
log("debug", file_name .. "\n has already been embedded, skipping.")
|
||||
-- add file reference if it was missing
|
||||
if not embeddings.files[file_name] then
|
||||
embeddings.files[file_name] = { sha512sum }
|
||||
end
|
||||
return
|
||||
end
|
||||
|
||||
log("debug", "Generating embeddings for " .. file_name .. ".")
|
||||
local file_sums = memorize_file(data_source, file_name)
|
||||
embeddings.files[file_name] = file_sums
|
||||
end
|
||||
loop()
|
||||
log("info", "Finished " .. utility.leftpad(f, #tostring(#file_list), "0")
|
||||
.. "/" .. #file_list
|
||||
.. " (" .. utility.leftpad(math.floor(f / #file_list * 100), 3, "0")
|
||||
.. "%) ETA: " .. timing.estimate(f, #file_list))
|
||||
end
|
||||
|
||||
timing.mark("Finished generating embeddings from source \"" .. source_name .. "\"")
|
||||
end
|
||||
|
||||
utility.save_data(embeddings)
|
||||
if utility.path_exists(tmp_file_path) then
|
||||
os.execute("rm " .. tmp_file_path:enquote())
|
||||
end
|
||||
end
|
||||
|
||||
refresh_sources()
|
||||
timing.mark("Finished.")
|
||||
print("")
|
||||
timing.display()
|
||||
log("warning")
|
||||
@@ -108,7 +108,7 @@ local export_ordered_list_of_prompts = function()
|
||||
output[#output + 1] = "## Scoring\n\n```json\n" .. json.encode(item.scoring, { indent = true, }) .. "\n```\n"
|
||||
end
|
||||
|
||||
utility.write_file("PRIVATE_DATA/Ordered Synopses.md", table.concat(output, "\n"))
|
||||
utility.write_file("PRIVATE_DATA/Ordered Synopses.md", table.concat(output, "\n"), "\n")
|
||||
os.execute("pandoc \"PRIVATE_DATA/Ordered Synopses.md\" -o \"PRIVATE_DATA/Ordered Synopses.epub\"")
|
||||
end
|
||||
|
||||
|
||||
@@ -6,16 +6,9 @@ local json = utility.require("dkjson")
|
||||
|
||||
local file_name = arg[1]
|
||||
|
||||
print(utility.OS)
|
||||
|
||||
|
||||
local function strip_markdown(text)
|
||||
local tab = text:split("\n")
|
||||
table.remove(tab, 1) -- codeblock opening
|
||||
table.remove(tab, #tab) -- end of codeblock
|
||||
print(table.concat(tab, "\n"))
|
||||
end
|
||||
|
||||
|
||||
os.exit(0)
|
||||
|
||||
local prompt = [[Return JSON: category, tags (array), summary (short)]]
|
||||
-- local prompt = [[Return YAML: category, tags (array), summary (short)]]
|
||||
@@ -24,39 +17,6 @@ if prompt:sub(-1) ~= "\n" then
|
||||
prompt = prompt .. "\n\n"
|
||||
end
|
||||
|
||||
local file_contents = utility.open(file_name, "r", function(file)
|
||||
return file:read("*all")
|
||||
end)
|
||||
local file_contents = utility.read_file(file_name)
|
||||
prompt = prompt .. file_contents
|
||||
|
||||
local model = "gemma3:4b"
|
||||
-- local output = utility.capture_safe("ollama run " .. model .. " --nowordwrap " .. prompt:enquote())
|
||||
|
||||
-- local embedding_model = "nomic-embed-text"
|
||||
-- local output = utility.capture_safe("ollama run " .. embedding_model .. " " .. file_contents:enquote())
|
||||
|
||||
-- output = output:sub(1, -2) -- strip extra newline from utility.capture_safe
|
||||
|
||||
-- strip YAML frontmatter (if present)
|
||||
-- can error, will return nil & error message
|
||||
local function strip_frontmatter(text)
|
||||
local tab = text:split("\n")
|
||||
if tab[1] == "---" then
|
||||
table.remove(tab, 1)
|
||||
while true do
|
||||
local done = tab[1] == "---"
|
||||
table.remove(tab, 1)
|
||||
if done then
|
||||
return table.concat(tab, "\n")
|
||||
elseif #tab < 1 then
|
||||
return nil, "Invalid YAML frontmatter."
|
||||
end
|
||||
end
|
||||
end
|
||||
return text
|
||||
end
|
||||
|
||||
-- print(output)
|
||||
|
||||
print(strip_frontmatter(file_contents))
|
||||
-- print(strip_markdown(output))
|
||||
-- print(utility.llm_prompt(prompt))
|
||||
|
||||
Reference in New Issue
Block a user