Mortdecai
df5542f7d6
Three-arm harness under scripts/native-bakeoff/: - arm A: /api/chat with JSON tools (current default) - arm B: /api/generate raw:true with canonical HF jinja template rendered directly - arm C: google-deepmind/gemma JAX ToolSampler (env-gated, JAX required) Interim finding from A+B sweep on matt-strix gemma4:26b Q4: Ollama's bidirectional JSON↔native tool-call translator is faithful. The "long" multi-tool task produces identical behavior (7 steps / 6 tools) on both arms. Earlier arm-B parser bug that looked like a divergence was a harness issue: preserving the model's <|channel>thought\n<channel|> prefix as assistant content tripped the jinja template's tool_response-following conditional, appending a spurious <turn|>\n that corrupted the next step's prompt. Fixed by dropping the channel prefix on the assistant message. Arm C left as scaffolded-but-not-run — the JAX/bf16 reference path would answer "does the GGUF runtime diverge from DeepMind's implementation" but requires a separate env with the `gemma` PyPI package. Parked pending SDXL eviction or vast-h100 session. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
82 lines
2.2 KiB
JSON
82 lines
2.2 KiB
JSON
{
|
|
"arm": "ollama-json",
|
|
"model": "gemma4:26b",
|
|
"num_ctx": 8192,
|
|
"num_predict": 2048,
|
|
"started_at": 1776600292.960776,
|
|
"turns": [
|
|
{
|
|
"step": 1,
|
|
"elapsed_s": 0.84,
|
|
"prompt_eval_count": 1393,
|
|
"eval_count": 24,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 3009
|
|
},
|
|
{
|
|
"step": 2,
|
|
"elapsed_s": 0.96,
|
|
"prompt_eval_count": 1507,
|
|
"eval_count": 29,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 3281
|
|
},
|
|
{
|
|
"step": 3,
|
|
"elapsed_s": 0.81,
|
|
"prompt_eval_count": 1645,
|
|
"eval_count": 24,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 3572
|
|
},
|
|
{
|
|
"step": 4,
|
|
"elapsed_s": 0.88,
|
|
"prompt_eval_count": 1808,
|
|
"eval_count": 25,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 4188
|
|
},
|
|
{
|
|
"step": 5,
|
|
"elapsed_s": 0.87,
|
|
"prompt_eval_count": 1923,
|
|
"eval_count": 24,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 4616
|
|
},
|
|
{
|
|
"step": 6,
|
|
"elapsed_s": 4.01,
|
|
"prompt_eval_count": 2053,
|
|
"eval_count": 177,
|
|
"content_len": 0,
|
|
"tool_call_count": 1,
|
|
"history_chars_before_append": 4843
|
|
},
|
|
{
|
|
"step": 7,
|
|
"elapsed_s": 5.56,
|
|
"prompt_eval_count": 2277,
|
|
"eval_count": 247,
|
|
"content_len": 950,
|
|
"tool_call_count": 0,
|
|
"history_chars_before_append": 4958
|
|
}
|
|
],
|
|
"final": {
|
|
"halt_reason": "no_tool_calls",
|
|
"steps_used": 7,
|
|
"tool_calls_total": 6,
|
|
"wall_clock_s": 13.93,
|
|
"final_message_count": 26,
|
|
"final_history_chars": 5908
|
|
},
|
|
"task": "long",
|
|
"task_prompt": "Research question with multiple steps: (1) check memory for what I have on home_automation, (2) search sethflix for any home-automation documentaries, (3) web_search for current news about Home Assistant version releases, (4) fetch the top search result for details, (5) check chat_search for prior mentions, (6) summarize all findings and write a new memory entry with the summary. Do each step in order and report back at the end."
|
|
} |