Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76263.26218684515,
-      "required_pages_per_second_upper_estimate": 0.006610306278884846
+      "required_token_positions_per_second": 76091.32222312325,
+      "required_pages_per_second_upper_estimate": 0.006595402958607285
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9532.907773355644,
-      "required_pages_per_second_upper_estimate": 0.003802819440464195
+      "required_token_positions_per_second": 9511.415277890406,
+      "required_pages_per_second_upper_estimate": 0.003794245762681668
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19065.81554671129,
-      "required_pages_per_second_upper_estimate": 0.004410540879838384
+      "required_token_positions_per_second": 19022.830555780813,
+      "required_pages_per_second_upper_estimate": 0.0044005970586966205
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28598.723320066933,
-      "required_pages_per_second_upper_estimate": 0.0017415122042869745
+      "required_token_positions_per_second": 28534.24583367122,
+      "required_pages_per_second_upper_estimate": 0.0017375858636527023
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28598.723320066933,
-      "required_pages_per_second_upper_estimate": 0.002201608887511773
+      "required_token_positions_per_second": 28534.24583367122,
+      "required_pages_per_second_upper_estimate": 0.0021966452321239252
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28598.723320066933,
-      "required_pages_per_second_upper_estimate": 0.3551004298653654
+      "required_token_positions_per_second": 28534.24583367122,
+      "required_pages_per_second_upper_estimate": 0.3542998352765961
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.043841725585936686,
-    "producer_p95_seconds": 0.11820209600000453,
-    "gpu_update_seconds": 0.9667564419073268,
-    "producer_batches_per_second": 22.80932118056901,
-    "gpu_batches_per_second": 1.0343866941575135,
-    "producer_headroom": 22.051058186848323,
+    "producer_batch_seconds": 0.05040427049999252,
+    "producer_p95_seconds": 0.12210148000008303,
+    "gpu_update_seconds": 0.9689409757371115,
+    "producer_batches_per_second": 19.83958879040117,
+    "gpu_batches_per_second": 1.0320546091460945,
+    "producer_headroom": 19.223390520794368,
     "sustainable": true,
-    "buffer_seconds": 60.905655840161586,
+    "buffer_seconds": 60.07434049570091,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -5,3 +5,4 @@
 {"step": 355500, "loss": 0.5927569717168808, "lr": 0.0003, "grad_norm": 0.14891977529208586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52035227901, "tokens_per_second": 151444.71984551175, "data_wait_s": 1.3001000070289592e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355600, "loss": 0.5656155925244093, "lr": 0.0003, "grad_norm": 0.16499899951484231, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52049949793, "tokens_per_second": 151893.7240686453, "data_wait_s": 1.3086000080875237e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355700, "loss": 0.5800438262522221, "lr": 0.0003, "grad_norm": 0.14151087830719344, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52064543552, "tokens_per_second": 150241.3835507587, "data_wait_s": 1.281000004382804e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355800, "loss": 0.6531054731458426, "lr": 0.0003, "grad_norm": 0.17719566197741718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52079015267, "tokens_per_second": 148980.5927603168, "data_wait_s": 1.1954000001423992e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1657939d3f6c1b68dcda821d72ef2872282f6d48ebf93388aef14785c6f048a4
-size 14991
+oid sha256:05c5a9c23f1a4e71721f425546e183c073896b55d54c18c1ab2695286ac4c6f4
+size 17109