bugfix

llmage/accounting.py

@@ -169,13 +169,8 @@ where a.llmid = b.id
 		# debug(f'{sql=}, {ns=}, {len(recs)=}')
 		for r in recs:
 			if r.usages is None:
-				io = json.loads(r.ioinfo)
+				output = await get_lastoutput(r.ioinfo)
-				if len(io['output']) == 0:
+				r.usages = output.get('usage')
-					llmusage.accounting_status = 'failed'
-					await sor.U('llmusage', {'id': llmusage.id, 'accounting_status': 'failed'})
-					debug(f'{len(io["output"])} is 0')
-					continue
-				r.usages = json.dumps(io['output'][-1].get('usage'))
 			if r.usages is None:
 				llmusage.accounting_status = 'failed'
 				await sor.U('llmusage', {'id': llmusage.id, 'accounting_status': 'failed'})

llmage/asyncinference.py

@@ -35,11 +35,8 @@ async def get_asynctask_status(taskid):
 		recs = await sor.R('llmusage', {'taskid': taskid})
 		if recs:
 			r = recs[0]
-			io = json.loads(r.ioinfo)
+			output = await get_lastoutput(r.ioinfo)
-			d = io.get('output', {})
+			return output
-			if isinstance(d, list):
-				return d[-1]
-			return d
 		return {
 			'taskid': taskid,
 			'status': 'FAILED',
@@ -129,14 +126,6 @@ async def modify_llmusage_status(llmusage):
 			'status': llmusage.status
 		})
 
-def get_llmusage_last_output(r):
-	io = json.loads(r.ioinfo)
-	outs = io.get('output', [])
-	if len(outs) == 0:
-		return None
-	d = DictObject(**outs[-1])
-	return d
-
 async def get_llm_llmusage(luid):
 	env = request._run_ns
 	async with get_sor_context(env, 'llmage') as sor:
@@ -165,7 +154,7 @@ async def query_task_status(request, upappid, apiname, luid, userid, taskid):
 
 	for apiname in apinames:
 		while True
-			lastoutout = get_llmusage_last_output(llmusage)
+			lastoutout = await get_lastoutput(llmusage.ioinfo)
 			if lastoutout['status'] in ['FAILED', 'SUCCEEDED']
 				return
 			ns = {'taskid': taskid}

llmage/llmclient.py

@@ -84,10 +84,13 @@ async def uapi_request(request, llm, callerid, callerorgid, params_kw=None):
 		llmusage.use_time = timestampstr()
 		llmusage.userid = callerid
 		llmusage.usages = json.dumps(usage, ensure_ascii=False)
-		llmusage.ioinfo = json.dumps({
+        ioinfo = json.dumps({
             "input": params_kw,
-			"output": outlines
+            'output': [d]
         }, ensure_ascii=False)
+        webpath = await write_llmio(llmusage.id, ioinfo)
+        llmusage.ioinfo = webpath
+		debug(f'webpath={webpath}:')
 		llmusage.transno = params_kw.transno
 		llmusage.responsed_seconds = responsed_seconds
 		llmusage.finish_seconds = finish_seconds

llmage/syncinference.py

@@ -50,10 +50,12 @@ async def sync_uapi_request(request, llm, callerid, callerorgid, params_kw=None)
 		llmusage.use_time = timestampstr()
 		llmusage.userid = callerid
 		llmusage.usages = json.dumps(usage, ensure_ascii=False)
-		llmusage.ioinfo = json.dumps({
+        ioinfo = json.dumps({
             "input": params_kw,
-			"output": [d]
+            'output': [d]
         }, ensure_ascii=False)
+        webpath = await write_llmio(llmusage.id, ioinfo)
+        llmusage.ioinfo = webpath
 		llmusage.transno = params_kw.transno
 		llmusage.responsed_seconds = responsed_seconds
 		llmusage.finish_seconds = finish_seconds

llmage/utils.py

@@ -25,6 +25,11 @@ async def append_new_llmoutput(webpath, output):
 		iostr = json.dumps(io, ensure_ascii=False, indent=4)
 		f.write(iostr.encode('utf-8'))
 	
+async def get_usage_from_lastoutput(webpath):
+	bin =  await read_webpath(webpath)
+	io = json.load(bin.decode('utf-8'))
+	return io['output'][-1]
+
 async def read_webpath(webpath):
 	fs = FileStorage()
 	p = fs.realPath(webpath)