Update funpack.py
This commit is contained in:
aimfordeb
+32
-26
@@ -201,7 +201,7 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
"dynamic_sim_threshold": 0.82,
|
||||
"last_feedback_tid": None,
|
||||
"last_feedback_word": None,
|
||||
"feedbacked_words": []
|
||||
"feedbacked_words": [] # global history of asked words
|
||||
},
|
||||
"prompt_histories": {
|
||||
prompt_key: {
|
||||
@@ -284,7 +284,7 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
|
||||
# ====================== STRICT WORD GROUPING ======================
|
||||
word_groups = [] # (start, end, full_word, word_token_list)
|
||||
current_words = set()
|
||||
current_words_set = set()
|
||||
if tokenizer and positive_prompt and full_token_ids:
|
||||
try:
|
||||
words = positive_prompt.split()
|
||||
@@ -296,18 +296,18 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
word_len = len(word_tokens)
|
||||
if token_idx + word_len <= len(full_token_ids) and full_token_ids[token_idx:token_idx + word_len] == word_tokens:
|
||||
word_groups.append((token_idx, token_idx + word_len, word, word_tokens))
|
||||
current_words.add(word.lower())
|
||||
current_words_set.add(word.lower())
|
||||
token_idx += word_len
|
||||
else:
|
||||
if token_idx < len(full_token_ids):
|
||||
single = full_token_ids[token_idx]
|
||||
decoded = tokenizer.decode([single]).strip()
|
||||
word_groups.append((token_idx, token_idx + 1, decoded, [single]))
|
||||
current_words.add(decoded.lower())
|
||||
current_words_set.add(decoded.lower())
|
||||
token_idx += 1
|
||||
except Exception:
|
||||
word_groups = [(i, i+1, tokenizer.decode([tid]).strip(), [tid]) for i, tid in enumerate(full_token_ids)]
|
||||
current_words = {w.lower() for _, _, w, _ in word_groups if self._is_valuable_token(w)}
|
||||
current_words_set = {w.lower() for _, _, w, _ in word_groups if self._is_valuable_token(w)}
|
||||
|
||||
# ====================== CROSS-PROMPT TRANSFER ======================
|
||||
if is_new_prompt and tokenizer and full_token_ids:
|
||||
@@ -443,7 +443,7 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
if full_token_ids[i] != prev_token_ids[i]:
|
||||
new_token_mask[i] = 0.0
|
||||
|
||||
# ====================== WORD-LEVEL IMPORTANCE ======================
|
||||
# Word-level importance update
|
||||
for start, end, full_word, word_token_list in word_groups:
|
||||
is_valuable = self._is_valuable_token(full_word)
|
||||
group_key = str(word_token_list[0])
|
||||
@@ -468,7 +468,7 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
else:
|
||||
token_importance[tid_str] = max(0.3, min(2.5, 1.0 + group_delta))
|
||||
|
||||
# ====================== INFUSION (only current prompt words) ======================
|
||||
# Infusion (only current prompt)
|
||||
if random.random() < 0.18 and token_library:
|
||||
infusion_weights = torch.ones(embed_seq_len, device=device)
|
||||
infusion_any = False
|
||||
@@ -661,25 +661,28 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
data["prompt_histories"] = prompt_histories
|
||||
data["global_adaptive"] = global_adaptive
|
||||
|
||||
# ====================== NEW FEEDBACK QUESTION (new words first + only current prompt) ======================
|
||||
# ====================== INTELLIGENT FEEDBACK QUESTION ======================
|
||||
if feedback_enabled and data.get("pending_feedback") is None and tokenizer and word_groups:
|
||||
feedbacked = set(global_adaptive.get("feedbacked_words", []))
|
||||
last_word = global_adaptive.get("last_feedback_word")
|
||||
|
||||
# Priority 1: New words in current prompt (never asked before)
|
||||
new_words = [ (full_word, group_key, score, best_tid)
|
||||
for start, end, full_word, word_token_list in word_groups
|
||||
for group_key in [str(word_token_list[0])]
|
||||
for best_tid in [word_token_list[0]]
|
||||
for score in [token_library.get(group_key, {}).get("fqc", 1) * (1.0 - abs(token_importance.get(group_key, 1.0) - 1.0))]
|
||||
if self._is_valuable_token(full_word) and full_word.lower() not in feedbacked and full_word != last_word ]
|
||||
# Priority 1: New words never asked before (highest priority when prompt changes)
|
||||
new_unasked = []
|
||||
for start, end, full_word, word_token_list in word_groups:
|
||||
if not self._is_valuable_token(full_word):
|
||||
continue
|
||||
if full_word.lower() in feedbacked or full_word == last_word:
|
||||
continue
|
||||
group_key = str(word_token_list[0])
|
||||
imp = token_importance.get(group_key, 1.0)
|
||||
freq = token_library.get(group_key, {}).get("fqc", 1)
|
||||
score = freq * (1.0 - abs(imp - 1.0))
|
||||
new_unasked.append((full_word, group_key, score, word_token_list[0]))
|
||||
|
||||
if new_words:
|
||||
# Pick the most "interesting" new word
|
||||
chosen = max(new_words, key=lambda x: x[2])
|
||||
full_word, group_key, score, best_tid = chosen
|
||||
if new_unasked:
|
||||
chosen = max(new_unasked, key=lambda x: x[2])
|
||||
else:
|
||||
# Priority 2: Existing words that are valuable and not recently asked
|
||||
# Priority 2: Re-ask words that might be problematic based on recent rating
|
||||
candidates = []
|
||||
for start, end, full_word, word_token_list in word_groups:
|
||||
if not self._is_valuable_token(full_word) or full_word == last_word:
|
||||
@@ -688,15 +691,17 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
imp = token_importance.get(group_key, 1.0)
|
||||
freq = token_library.get(group_key, {}).get("fqc", 1)
|
||||
score = freq * (1.0 - abs(imp - 1.0))
|
||||
candidates.append((full_word, group_key, score, word_token_list[0]))
|
||||
# Bonus if recent rating was very high or very low
|
||||
rating_bonus = 1.5 if rating >= 8 or rating <= 3 else 1.0
|
||||
candidates.append((full_word, group_key, score * rating_bonus, word_token_list[0]))
|
||||
|
||||
if candidates:
|
||||
chosen = max(candidates, key=lambda x: x[2])
|
||||
full_word, group_key, score, best_tid = chosen
|
||||
else:
|
||||
full_word = None
|
||||
chosen = None
|
||||
|
||||
if full_word:
|
||||
if chosen:
|
||||
full_word, group_key, _, best_tid = chosen
|
||||
data["pending_feedback"] = {
|
||||
"token_id": int(best_tid),
|
||||
"token_text": full_word,
|
||||
@@ -705,11 +710,12 @@ class FunPackGemmaEmbeddingRefiner:
|
||||
}
|
||||
global_adaptive["last_feedback_tid"] = int(best_tid)
|
||||
global_adaptive["last_feedback_word"] = full_word
|
||||
global_adaptive.setdefault("feedbacked_words", []).append(full_word)
|
||||
if full_word.lower() not in feedbacked:
|
||||
global_adaptive.setdefault("feedbacked_words", []).append(full_word)
|
||||
|
||||
feedback_question_output = f"Word: '{full_word}' – How well represented? (1=absent, 5=perfect, 6=overrepresented)"
|
||||
else:
|
||||
feedback_question_output = "Feedback enabled but no new or suitable word found in current prompt."
|
||||
feedback_question_output = "Feedback enabled but no suitable word found in current prompt."
|
||||
|
||||
# ====================== FINAL SAVE ======================
|
||||
with open(json_file, "w", encoding="utf-8") as f:
|
||||
|
||||
Reference in New Issue
Block a user