emilandClaude Opus 4.7
75c9ee4576
C++ core (libmemba.so): - include/memba/state.h — C API (state_new/free/save/load/get_size) - src/state.cpp — MEMB file format: magic, version, SHA-256 model_id, CRC-32, opaque llama_state_*_data() blob - src/cli.cpp — minimal demo binary with greedy sampler - CMakeLists.txt + build.sh with llama.cpp submodule, CUDA auto-detect Python SDK (memba): - core.py — file I/O via llama-cpp-python's exposed C functions, unwraps _LlamaContext to access raw context pointer (≥0.3.x) - session.py — high-level Session with auto-save/load, ChatML wrapper for instruct models, raw mode for base models - cli.py — typer-based: chat (REPL), run (one-shot), list, rm, info Examples: - 01_basic_save_load.py, 02_chat_session.py Experiments (throwaway POCs documenting product-direction findings): - recall_poc.py — git log → state → cross-process query - mood_poc.py — batch sentiment trajectory, Mamba vs Transformer - mood_stream_poc.py, mood_batch_poc.py — variants - diag_saveload.py — minimal save/load isolation test - README.md documents the headline finding: save/load is byte-identical, but Falcon-Mamba-7B-Instruct does not retain facts across conversation turns even in-process — limits viable products to single-prompt analysis and persona priming. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
74 lines
3.0 KiB
Python
74 lines
3.0 KiB
Python
"""
|
|
mood_poc.py — sentiment-trajectory test on Mamba vs Transformer.
|
|
|
|
A chat log is constructed with a deliberate emotional arc:
|
|
msg 1-5 : optimistic / energetic
|
|
msg 6-10 : frustrated, hitting friction
|
|
msg 11-15 : burnt out, angry
|
|
|
|
Both models see the same prompt and answer 3 questions:
|
|
Q1. Current mood at message 15
|
|
Q2. Trajectory from start to end
|
|
Q3. Approximate message number where mood shifted
|
|
|
|
Pass criterion: model identifies negative trend AND points at a shift
|
|
between msgs 6-11. Generic "they seem fine" or "they were happy throughout"
|
|
counts as failure.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
from llama_cpp import Llama
|
|
|
|
MAMBA = "/home/emil/Desktop/Coding/AI/Memba/falcon-mamba-7B-instruct-Q4_K_M.gguf"
|
|
GEMMA = "/home/emil/Programs/Llama.cpp/llama.cpp/models/gemma-4-E4B-it-Q8_0.gguf"
|
|
|
|
CHAT_LOG = [
|
|
"Morning team! Coffee in hand, ready to tackle the auth refactor today.",
|
|
"Just pushed PR #234 fixing the token validation bug. Should be a quick merge.",
|
|
"Code review comments came in fast, all good catches. Iterating now.",
|
|
"Basic flow working locally, tests passing. Feeling good about this.",
|
|
"Heading to lunch, hopefully wrap this up by EOD.",
|
|
"Back. CI is failing on something unrelated, looking into it.",
|
|
"OK the 'unrelated' thing is actually related. Auth tests use a stale fixture.",
|
|
"Why does the fixture rebuild take 12 minutes. Every. Single. Time.",
|
|
"Cancelled the run twice now. Going to bypass and run tests locally.",
|
|
"Local passes, CI fails. Classic.",
|
|
"Two hours gone on this fixture issue. Not even what I was supposed to be doing.",
|
|
"Now there's a merge conflict with main because someone restructured migrations.",
|
|
"Whoever shipped those migrations on a Friday afternoon, I will find you.",
|
|
"Closing the laptop. Will fight this tomorrow.",
|
|
"Actually no. One more try before I sleep.",
|
|
]
|
|
|
|
PROMPT = """You are observing a person's chat messages from one workday. Here they are in order:
|
|
|
|
""" + "\n".join(f"[msg {i+1:>2}] {m}" for i, m in enumerate(CHAT_LOG)) + """
|
|
|
|
Now answer these THREE questions, briefly and directly:
|
|
|
|
Q1: What is this person's mood at message 15 (the last one)? One short sentence.
|
|
Q2: How did their mood change from message 1 to message 15? One short sentence.
|
|
Q3: Around which message number does the mood clearly shift from positive to negative? Just give the number.
|
|
|
|
Format your answer as:
|
|
A1: ...
|
|
A2: ...
|
|
A3: ..."""
|
|
|
|
|
|
def run(label: str, model_path: str) -> None:
|
|
print(f"\n{'='*60}\n {label}\n{'='*60}")
|
|
llm = Llama(model_path=model_path, n_ctx=4096, n_gpu_layers=-1, verbose=False)
|
|
out = llm.create_chat_completion(
|
|
messages=[{"role": "user", "content": PROMPT}],
|
|
max_tokens=300,
|
|
temperature=0.3, # low temp so we test capability, not creativity
|
|
)
|
|
print(out["choices"][0]["message"]["content"])
|
|
del llm # free GPU memory before loading next
|
|
|
|
|
|
if __name__ == "__main__":
|
|
run("FALCON-MAMBA-7B-INSTRUCT (SSM)", MAMBA)
|
|
run("GEMMA-4-E4B-IT (Transformer)", GEMMA)
|