fix(pi052): port the smolvla2 text-head fixes to pi052

pi052 had the same text-CE collapse bug smolvla2 had — PaliGemma's
embed_prefix flags the language block att=0, so make_att_2d_masks makes
it fully bidirectional and the text cross-entropy degenerates into a
copy task. Ported the three model-specific fixes:

- _mark_target_span_causal: set att=1 on supervised target language
  positions so the text-CE is genuine causal next-token prediction.
  Applied in both _compute_all_losses_fused and _compute_text_and_fast_loss.
- flow_loss_weight 10.0 -> 5.0: the paper's a=10 swamps the LM head once
  the flow-only low_level recipe fires often (matches SmolVLA2Config).
- _flatten_say_tool_calls in the text tokenizer: serialize `say` tool
  calls into a <say>...</say> marker so the spoken reply is tokenized
  and supervised (PaliGemma's flat prompt has no structured calls, so
  they were dropped entirely).

select_message needed no change: pi052's prefix is [images, language]
with no trailing state token, so it already decodes from the last
language token.

Regression tests mirror the smolvla2 attention-masking + tool-call suite.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/lerobot/policies/pi052/configuration_pi052.py

@@ -49,11 +49,9 @@ class PI052Config(PI05Config):
     """π0.5 with the PaliGemma LM head re-enabled for subtask prediction.
 
     See ``SmolVLA2Config`` for the analogous SmolVLM2-backed dual-head
-    config. Same recipe-driven training surface; the only differences
-    are which backbone the policy uses (PaliGemma here vs SmolVLM2
-    there) and the default loss-weight scale (paper §IV.D uses
-    ``α=10`` between the two heads, which we encode as
-    ``flow_loss_weight=10, text_loss_weight=1``).
+    config. Same recipe-driven training surface; the only difference is
+    which backbone the policy uses (PaliGemma here vs SmolVLM2 there).
+    The flow:text loss split is the milder 5:1 (see ``flow_loss_weight``).
     """
 
     # Recipe / language stack ---------------------------------------------
@@ -72,16 +70,20 @@ class PI052Config(PI05Config):
     samples text auto-regressively after the prefix."""
 
     # Loss weights --------------------------------------------------------
-    # Paper §IV.D: total = H(text) + α * MSE(flow), α = 10. We split
-    # the same total into two configurable knobs so individual scaling
-    # is recoverable.
+    # Paper §IV.D uses α=10 between the flow and text terms, assuming
+    # text is a rare auxiliary task. With the recipe stack the flow-only
+    # `low_level` branch fires on a large share of samples, so α=10
+    # swamps the LM head and collapses generation into degenerate
+    # repetition. We use the milder 5:1 split (matches SmolVLA2Config).
     text_loss_weight: float = 1.0
     """Weight on the LM-head cross-entropy term. Set to ``0`` to disable
     text training entirely (reverts to flow-only / π0.5 behaviour)."""
 
-    flow_loss_weight: float = 10.0
-    """Weight on the action-expert flow-matching term. Default ``10.0``
-    matches the paper's α."""
+    flow_loss_weight: float = 5.0
+    """Weight on the action-expert flow-matching term. ``5.0`` — a milder
+    flow:text split than the paper's α=10, since the flow-only
+    ``low_level`` recipe already gives the action expert frequent
+    gradient. Lower it further if the LM head still underfits."""
 
     # Backbone training ---------------------------------------------------
     unfreeze_lm_head: bool = True

src/lerobot/policies/pi052/modeling_pi052.py

@@ -86,6 +86,41 @@ def _shifted_ce(logits: Tensor, labels: Tensor) -> Tensor:
     return loss / valid.sum().clamp(min=1)
 
 
+def _mark_target_span_causal(
+    prefix_att_masks: Tensor, text_labels: Tensor, lang_start: int, lang_end: int
+) -> Tensor:
+    """Make the supervised text-target span causally masked.
+
+    ``embed_prefix`` lays the PaliGemma prefix out as ``[images,
+    language]`` with the language block flagged ``att=0`` — which
+    ``make_att_2d_masks`` turns into one fully *bidirectional* block.
+    A supervised target token's hidden state then attends to the very
+    tokens it is trained to predict, so the text cross-entropy
+    degenerates into a copy task (loss → ~0) and the LM head never
+    learns causal next-token prediction. At inference ``select_message``
+    decodes autoregressively (causally) and the head collapses to
+    repeated/garbage tokens.
+
+    Fix: set ``att=1`` on the language positions that are supervised
+    targets (``text_labels != -100``). Under ``make_att_2d_masks``'s
+    cumulative-block rule each target token then attends bidirectionally
+    to images + the user prompt and causally to *earlier* targets only —
+    genuine next-token prediction, matching inference. Non-target
+    language (the user prompt, the flow-only ``low_level`` subtask) stays
+    ``att=0`` / bidirectional. The action expert / FAST tokens are
+    unaffected: they sit at a strictly higher cumsum and still attend to
+    every prefix token.
+    """
+    att = prefix_att_masks.clone()
+    n = min(text_labels.shape[1], lang_end - lang_start)
+    if n <= 0:
+        return att
+    target = text_labels[:, :n] != -100  # (B, n) bool
+    seg = att[:, lang_start : lang_start + n].bool()
+    att[:, lang_start : lang_start + n] = seg | target
+    return att
+
+
 def _fast_ce(
     fast_logits: Tensor,
     action_tokens: Tensor,
@@ -519,6 +554,16 @@ class PI052Policy(PI05Policy):
         )
         non_fast_prefix_len = prefix_embs.shape[1]  # images + language only
 
+        # Causal-mask the supervised text-target span so the text-CE is
+        # genuine next-token prediction, not a bidirectional copy task
+        # (see ``_mark_target_span_causal``).
+        if text_labels is not None:
+            lang_start = non_fast_prefix_len - text_labels.shape[1]
+            if lang_start >= 0:
+                prefix_att = _mark_target_span_causal(
+                    prefix_att, text_labels, lang_start, non_fast_prefix_len
+                )
+
         fast_len = 0
         if action_tokens is not None and action_mask is not None:
             emb_dim = prefix_embs.shape[-1]
@@ -640,6 +685,16 @@ class PI052Policy(PI05Policy):
             images, img_masks, lang_tokens, lang_masks
         )
 
+        # Causal-mask the supervised text-target span (see
+        # ``_mark_target_span_causal``) before the FAST tokens are
+        # appended — same fix as ``_compute_all_losses_fused``.
+        if text_labels is not None:
+            lang_start = prefix_embs.shape[1] - text_labels.shape[1]
+            if lang_start >= 0:
+                prefix_att = _mark_target_span_causal(
+                    prefix_att, text_labels, lang_start, prefix_embs.shape[1]
+                )
+
         fast_len = 0
         if action_tokens is not None and action_mask is not None:
             emb_dim = prefix_embs.shape[-1]

src/lerobot/policies/pi052/text_processor_pi052.py

@@ -125,6 +125,62 @@ def _dump_recipe_sample(
     print("==============================================\n", flush=True)
 
 
+def _content_to_text(content: Any) -> str:
+    """Collapse a message's ``content`` (string or multimodal blocks) to text."""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = [
+            b["text"]
+            for b in content
+            if isinstance(b, dict) and b.get("type") == "text" and isinstance(b.get("text"), str)
+        ]
+        return "\n".join(parts)
+    return ""
+
+
+def _flatten_say_tool_calls(message: dict[str, Any]) -> dict[str, Any]:
+    """Serialize assistant ``say`` tool calls into a ``<say>...</say>`` marker.
+
+    PaliGemma's flat text prompt has no notion of structured tool calls,
+    and ``_format_messages`` only reads ``role`` / ``content`` — so
+    without this a ``say`` tool call is dropped entirely and never
+    supervised. Rewriting it into the content text as a ``<say>...</say>``
+    marker lets the LM head learn to emit it; the runtime parses it back
+    via ``_split_plan_and_say``. Messages without ``say`` tool calls are
+    returned unchanged (the structured calls, if any, are still dropped).
+    """
+    tool_calls = message.get("tool_calls")
+    if not tool_calls:
+        return message
+    say_texts: list[str] = []
+    for call in tool_calls:
+        if not isinstance(call, dict):
+            continue
+        fn = call.get("function") or {}
+        if fn.get("name") != "say":
+            continue
+        args = fn.get("arguments")
+        if isinstance(args, str):
+            try:
+                import json  # noqa: PLC0415
+
+                args = json.loads(args)
+            except (ValueError, TypeError):
+                args = {}
+        text = args.get("text", "") if isinstance(args, dict) else ""
+        if text:
+            say_texts.append(str(text))
+    new = dict(message)
+    new.pop("tool_calls", None)
+    if not say_texts:
+        return new
+    base = _content_to_text(new.get("content")).strip()
+    marker = "".join(f"<say>{t}</say>" for t in say_texts)
+    new["content"] = f"{base}\n{marker}" if base else marker
+    return new
+
+
 def _strip_blocks(message: dict[str, Any]) -> dict[str, Any]:
     """Normalise a message's content to a plain string.
 
@@ -253,7 +309,10 @@ class PI052TextTokenizerStep(ProcessorStep):
                 complementary,
             )
 
-        messages = [_strip_blocks(m) for m in messages]
+        # Flatten ``say`` tool calls into ``<say>...</say>`` text before
+        # stripping, so the spoken reply is actually tokenized and
+        # supervised (PaliGemma's flat prompt has no structured calls).
+        messages = [_strip_blocks(_flatten_say_tool_calls(m)) for m in messages]
         prompt, spans = _format_messages(messages)
 
         tokenizer = self._ensure_tokenizer()

tests/policies/pi052/test_pi052_attention_masking.py

@@ -0,0 +1,138 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Attention-masking tests for the PI052 (π0.5 v2) text head.
+
+Regression coverage for the text-CE collapse bug: PaliGemma's
+``embed_prefix`` flags every language token ``att=0``, which
+``make_att_2d_masks`` turns into one fully *bidirectional* block. Under
+that mask the text cross-entropy degenerates into a copy task — a
+supervised target token attends to the tokens it is trained to predict —
+and the LM head never learns causal generation, so ``select_message``
+collapses at inference.
+
+``_mark_target_span_causal`` sets ``att=1`` on the supervised target
+language positions so each target token attends causally among the
+targets while staying bidirectional to images + the user prompt. These
+tests pin that behaviour for the PaliGemma prefix layout.
+"""
+
+import pytest
+import torch
+
+# modeling_pi052 / modeling_pi05 import transformers transitively.
+pytest.importorskip("transformers")
+
+from lerobot.policies.pi05.modeling_pi05 import make_att_2d_masks  # noqa: E402
+from lerobot.policies.pi052.modeling_pi052 import _mark_target_span_causal  # noqa: E402
+
+# ---------------------------------------------------------------------------
+# A synthetic PI052 prefix layout: [images, prompt-lang, target-lang]
+#
+#   indices 0-1  : 2 image tokens          (att = 0)
+#   indices 2-4  : 3 user-prompt lang      (att = 0)
+#   indices 5-8  : 4 supervised target lang(att = 0 from embed_prefix)
+#
+# ``text_labels`` covers the 7 language tokens; -100 on the prompt span,
+# real ids on the 4-token target span. PaliGemma's prefix has no state
+# token (unlike SmolVLA), so the lang span ends at the prefix end.
+# ---------------------------------------------------------------------------
+N_IMAGE = 2
+N_PROMPT = 3
+N_TARGET = 4
+LANG_START = N_IMAGE
+LANG_END = N_IMAGE + N_PROMPT + N_TARGET  # = prefix length
+PREFIX_LEN = LANG_END
+
+
+def _embed_prefix_att_masks() -> torch.Tensor:
+    """Mimic PaliGemma ``embed_prefix``: images + lang all att=0."""
+    return torch.zeros(1, PREFIX_LEN, dtype=torch.bool)
+
+
+def _text_labels() -> torch.Tensor:
+    """-100 over the prompt span, real ids over the target span."""
+    labels = torch.full((1, N_PROMPT + N_TARGET), -100, dtype=torch.long)
+    labels[0, N_PROMPT:] = torch.arange(10, 10 + N_TARGET)
+    return labels
+
+
+def _attends(prefix_att_masks: torch.Tensor) -> torch.Tensor:
+    """2D boolean attendance matrix; ``[i, j]`` True ⇒ i attends to j."""
+    pad = torch.ones(1, PREFIX_LEN, dtype=torch.bool)
+    return make_att_2d_masks(pad, prefix_att_masks)[0]
+
+
+def test_mark_sets_att_on_targets_only():
+    """Only the supervised target language positions flip to att=1."""
+    marked = _mark_target_span_causal(
+        _embed_prefix_att_masks(), _text_labels(), LANG_START, LANG_END
+    )
+    expected = [False] * PREFIX_LEN
+    for i in range(LANG_START + N_PROMPT, LANG_END):  # target span
+        expected[i] = True
+    assert marked[0].tolist() == expected
+
+
+def test_target_tokens_attend_causally_among_themselves():
+    """A target token must NOT attend to later targets, but must attend
+    to earlier ones — genuine causal next-token prediction."""
+    marked = _mark_target_span_causal(
+        _embed_prefix_att_masks(), _text_labels(), LANG_START, LANG_END
+    )
+    attends = _attends(marked)
+    tgt = range(LANG_START + N_PROMPT, LANG_END)
+    for i in tgt:
+        for j in tgt:
+            if j > i:
+                assert not attends[i, j], f"target {i} must not see future target {j}"
+            else:
+                assert attends[i, j], f"target {i} must see earlier/self target {j}"
+
+
+def test_target_tokens_attend_prompt_and_images_bidirectionally():
+    """Targets keep full visibility of images + the user prompt."""
+    marked = _mark_target_span_causal(
+        _embed_prefix_att_masks(), _text_labels(), LANG_START, LANG_END
+    )
+    attends = _attends(marked)
+    context = list(range(0, LANG_START + N_PROMPT))  # images + prompt
+    for i in range(LANG_START + N_PROMPT, LANG_END):
+        for j in context:
+            assert attends[i, j], f"target {i} must attend context {j}"
+
+
+def test_non_target_subtask_stays_bidirectional():
+    """A flow-only / non-target language span (all -100 labels) leaves the
+    mask untouched — the action expert reads it bidirectionally."""
+    all_ignored = torch.full((1, N_PROMPT + N_TARGET), -100, dtype=torch.long)
+    marked = _mark_target_span_causal(
+        _embed_prefix_att_masks(), all_ignored, LANG_START, LANG_END
+    )
+    assert torch.equal(marked, _embed_prefix_att_masks())
+
+
+def test_unmarked_mask_is_bidirectional_the_bug():
+    """Documents the bug the fix prevents: without ``_mark_target_span_causal``
+    a target token attends *bidirectionally* to later targets — the
+    text-CE can copy the answer it is trained to predict."""
+    attends = _attends(_embed_prefix_att_masks())
+    first_tgt = LANG_START + N_PROMPT
+    last_tgt = LANG_END - 1
+    assert attends[first_tgt, last_tgt], (
+        "raw embed_prefix mask is bidirectional over language — the first "
+        "target token can see the last, which is the collapse bug"
+    )

tests/policies/pi052/test_pi052_text_processor.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python
+
+# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for PI052's text-tokenizer ``say`` tool-call flattening.
+
+PaliGemma's flat prompt has no structured tool calls, so an assistant
+``say`` tool call must be serialized into a ``<say>...</say>`` text
+marker — otherwise the spoken reply is dropped and never supervised.
+"""
+
+from lerobot.policies.pi052.text_processor_pi052 import _flatten_say_tool_calls
+
+
+def _say_call(text):
+    return {"type": "function", "function": {"name": "say", "arguments": {"text": text}}}
+
+
+def test_flatten_appends_say_marker_and_drops_tool_calls():
+    msg = {"role": "assistant", "content": "Heading to the cube.", "tool_calls": [_say_call("On it!")]}
+    out = _flatten_say_tool_calls(msg)
+    assert "tool_calls" not in out
+    assert out["content"] == "Heading to the cube.\n<say>On it!</say>"
+
+
+def test_flatten_marker_only_when_content_empty_or_none():
+    out = _flatten_say_tool_calls({"role": "assistant", "tool_calls": [_say_call("hi")]})
+    assert out["content"] == "<say>hi</say>"
+
+
+def test_flatten_accepts_json_string_arguments():
+    call = {"type": "function", "function": {"name": "say", "arguments": '{"text": "hello there"}'}}
+    out = _flatten_say_tool_calls({"role": "assistant", "content": "p", "tool_calls": [call]})
+    assert out["content"] == "p\n<say>hello there</say>"
+
+
+def test_flatten_leaves_messages_without_tool_calls_untouched():
+    msg = {"role": "assistant", "content": "just a plan"}
+    assert _flatten_say_tool_calls(msg) == msg
+
+
+def test_flatten_drops_non_say_tool_calls_but_keeps_content():
+    weather = {"type": "function", "function": {"name": "check_weather", "arguments": {}}}
+    out = _flatten_say_tool_calls(
+        {"role": "assistant", "content": "plan only", "tool_calls": [weather]}
+    )
+    assert out["content"] == "plan only"
+    assert "tool_calls" not in out