Update refresh_sources.lua

This commit is contained in:
2026-09-15 08:14:35 -06:00
parent 692d4f5690
commit 7b4608ec7e
+63 -10
View File
@@ -3,6 +3,8 @@
package.path = (arg[0]:match("@?(.*/)") or arg[0]:match("@?(.*\\)")) .. "lib" .. package.config:sub(1, 1) .. "?.lua;" .. package.path package.path = (arg[0]:match("@?(.*/)") or arg[0]:match("@?(.*\\)")) .. "lib" .. package.config:sub(1, 1) .. "?.lua;" .. package.path
local utility = require "utility" local utility = require "utility"
local json = utility.require("dkjson")
local text_processing = utility.require("text_processing")
local timing = utility.require("timing") 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 make utility have a function for getting/setting defaults where locking is only used to set defaults if they aren't present
@@ -17,7 +19,9 @@ if not config.models then
utility.save_config() utility.save_config()
end end
local refresht_file_list = function(data_source) -- TODO recognize uninitialized source and initialize it
-- this would need to be in the config
local refresh_file_list = function(data_source)
timing.mark("Assembling file list.") timing.mark("Assembling file list.")
os.execute("cd " .. ("PRIVATE_DATA" .. utility.path_separator .. data_source.path):enquote() .. " && " .. data_source.refresh_command) os.execute("cd " .. ("PRIVATE_DATA" .. utility.path_separator .. data_source.path):enquote() .. " && " .. data_source.refresh_command)
@@ -32,9 +36,11 @@ local refresht_file_list = function(data_source)
file_list[#file_list + 1] = file_name file_list[#file_list + 1] = file_name
end) end)
timing.mark("Finished assembling file list.")
return file_list return file_list
end end
-- returns nothing when too much text is sent
local generate_embeddings = function(text) local generate_embeddings = function(text)
if #text > config.models.embedding.max_chunk_size then if #text > config.models.embedding.max_chunk_size then
return nil, "generate_embeddings() must only be passed appropriately-sized chunks!" return nil, "generate_embeddings() must only be passed appropriately-sized chunks!"
@@ -42,21 +48,68 @@ local generate_embeddings = function(text)
local result = utility.llm_prompt(text, config.models.embedding.model) local result = utility.llm_prompt(text, config.models.embedding.model)
-- skipping unnecessary decoding to an array and putting back into text for saving result = setmetatable({}, {
-- this would cause problems if we intended to use the results immediately
return setmetatable({}, {
__tojson = function() return result end, __tojson = function() return result end,
}) })
return json.decode(result)
end end
local process_file = function(file_name) -- returns nothing for empty files
local process_file = function(data_source, file_name)
local text = utility.read_file(file_name) local text = utility.read_file(file_name)
if w
if data_source.strip_frontmatter then
text = text_processing.strip_frontmatter(text)
end
if #text == 0 then
print(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)
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 embeddings = {}
for i = 1, #chunks do
embeddings[i] = generate_embeddings(chunks[i])
end
if embeddings[1] == nil then
-- TODO average all other embeddings to this one
end
return chunks, embeddings
end end
local to_be_named_generate_file_embeddings = function(text) -- TODO somewhere near the top of this function chain needs to be the ability to recognize a file as unchanged and skip re-generating embeddings
if #text > config.models.embedding.max_chunk_size then -- this means running the same shasum command I will put below on that source file before trying to generate an embedding
-- TODO, split, process each chunk, return both a total and each chunk -- something something merging new changes into the old data structure means we need to load that data structure first so we only replace modified chunks
local refresh_sources = function()
local sources = utility.load_data("PRIVATE_DATA/sources.json")
for _, data_source in pairs(sources) do
local file_list = refresh_file_list(data_source)
for _, file_name in ipairs(file_list) do
local chunks, embeddings = process_file(file_name)
-- TODO decide how these will be stored
-- for chunks, save to a specific local tmp file to shasum, then mv that file based on the sum
-- for embeddings, store sum = embedding ? (look at how generate_embeddings.lua worked)
end
end end
-- return just the file and its embedding
end end