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": 72106.31265299165,
-      "required_pages_per_second_upper_estimate": 0.006249992428982695
+      "required_token_positions_per_second": 74696.71348722196,
+      "required_pages_per_second_upper_estimate": 0.006474521807982888
     },
     {
       "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": 9013.289081623956,
-      "required_pages_per_second_upper_estimate": 0.003595535775340656
+      "required_token_positions_per_second": 9337.089185902745,
+      "required_pages_per_second_upper_estimate": 0.003724704478180447
     },
     {
       "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": 18026.578163247912,
-      "required_pages_per_second_upper_estimate": 0.004170131601127401
+      "required_token_positions_per_second": 18674.17837180549,
+      "required_pages_per_second_upper_estimate": 0.004319942512002795
     },
     {
       "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": 27039.867244871868,
-      "required_pages_per_second_upper_estimate": 0.0016465860479933394
+      "required_token_positions_per_second": 28011.267557708234,
+      "required_pages_per_second_upper_estimate": 0.001705739230501514
     },
     {
       "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": 27039.867244871868,
-      "required_pages_per_second_upper_estimate": 0.0020816038316534566
+      "required_token_positions_per_second": 28011.267557708234,
+      "required_pages_per_second_upper_estimate": 0.00215638491674377
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27039.867244871868,
-      "required_pages_per_second_upper_estimate": 0.3357446545671166
+      "required_token_positions_per_second": 28011.267557708234,
+      "required_pages_per_second_upper_estimate": 0.34780619538483226
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 59,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.36509881958590995,
-    "producer_p95_seconds": 0.31781539599978714,
-    "gpu_update_seconds": 1.0224902271013725,
-    "producer_batches_per_second": 2.738984478597291,
-    "gpu_batches_per_second": 0.9780044576414882,
-    "producer_headroom": 2.8005848615480784,
+    "producer_batch_seconds": 0.08802556801558126,
+    "producer_p95_seconds": 0.23038905000066734,
+    "gpu_update_seconds": 0.9870313773926925,
+    "producer_batches_per_second": 11.360335667734535,
+    "gpu_batches_per_second": 1.013139017567572,
+    "producer_headroom": 11.21300776176735,
     "sustainable": true,
-    "buffer_seconds": 60.32692339898098,
+    "buffer_seconds": 60.208914020954246,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355498.6706685716,
+    "generation_bytes_per_second": 356055.6574135698,
     "upload_bytes_per_second": 5775043.062551655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.28173913996564,
+    "estimated_full_shards_per_hour": 76.40125552945443,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

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

@@ -204,3 +204,4 @@
 {"step": 375400, "loss": 0.5743808448314667, "lr": 0.0003, "grad_norm": 0.15233685722282306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54947279592, "tokens_per_second": 147841.1089620428, "data_wait_s": 9.740000677993521e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 375500, "loss": 0.5590781159698963, "lr": 0.0003, "grad_norm": 0.15987953919798303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54961898171, "tokens_per_second": 148240.55917270106, "data_wait_s": 1.4318000467028469e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375600, "loss": 0.5565471760928631, "lr": 0.0003, "grad_norm": 0.15656504648731412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54976547112, "tokens_per_second": 148730.40929264852, "data_wait_s": 1.4162997103994712e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 375700, "loss": 0.5926074273884296, "lr": 0.0003, "grad_norm": 0.15002919995749728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54991231745, "tokens_per_second": 149483.97750365784, "data_wait_s": 1.0679999832063913e-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 sha256:174279524ee3202a8c007af816aafa946b84c7efe54001549e45a026a54bda04
-size 439605
+oid sha256:8d27db2f6a4fbe5989ba4f3c73722986c11ae4cafcdb0a3050e0da5858144265
+size 441723