Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77262.65640258745,
-      "required_pages_per_second_upper_estimate": 0.005345209843993
+      "required_token_positions_per_second": 77590.7020531781,
+      "required_pages_per_second_upper_estimate": 0.005367904803271639
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9657.832050323432,
-      "required_pages_per_second_upper_estimate": 0.004043532601003587
+      "required_token_positions_per_second": 9698.837756647263,
+      "required_pages_per_second_upper_estimate": 0.0040607008339448546
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19315.664100646864,
-      "required_pages_per_second_upper_estimate": 0.004022943443847527
+      "required_token_positions_per_second": 19397.675513294525,
+      "required_pages_per_second_upper_estimate": 0.004040024258315642
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28973.496150970295,
-      "required_pages_per_second_upper_estimate": 0.001736916854283624
+      "required_token_positions_per_second": 29096.513269941788,
+      "required_pages_per_second_upper_estimate": 0.0017442915427297002
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28973.496150970295,
-      "required_pages_per_second_upper_estimate": 0.002252561884796746
+      "required_token_positions_per_second": 29096.513269941788,
+      "required_pages_per_second_upper_estimate": 0.002262125924701658
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28973.496150970295,
-      "required_pages_per_second_upper_estimate": 0.3716408993082477
+      "required_token_positions_per_second": 29096.513269941788,
+      "required_pages_per_second_upper_estimate": 0.3732188308249226
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3597521858673076,
-    "producer_p95_seconds": 0.2005924570003117,
-    "gpu_update_seconds": 0.9542514253694612,
-    "producer_batches_per_second": 2.7796912410390298,
-    "gpu_batches_per_second": 1.0479418457382195,
-    "producer_headroom": 2.6525243288485005,
+    "producer_batch_seconds": 0.06030031553135018,
+    "producer_p95_seconds": 0.2028617980013223,
+    "gpu_update_seconds": 0.9502169467350516,
+    "producer_batches_per_second": 16.58366114983427,
+    "gpu_batches_per_second": 1.0523912496362047,
+    "producer_headroom": 15.758075863484216,
     "sustainable": true,
-    "buffer_seconds": 60.11783979827605,
+    "buffer_seconds": 60.813884591043305,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439039.2488571164,
+    "generation_bytes_per_second": 440619.55935219006,
     "upload_bytes_per_second": 5841925.60512766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.20760249409788,
+    "estimated_full_shards_per_hour": 94.54670033859516,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 495300, "loss": 0.5538405776023865, "lr": 0.0003, "grad_norm": 0.181019979305347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72495589433, "tokens_per_second": 151168.66815730848, "data_wait_s": 1.831599911383819e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495400, "loss": 0.5644538998603821, "lr": 0.0003, "grad_norm": 0.16119446364049433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72510205476, "tokens_per_second": 152302.51850732145, "data_wait_s": 1.0859999747481197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495500, "loss": 0.5806675553321838, "lr": 0.0003, "grad_norm": 0.17175863049179055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72524850839, "tokens_per_second": 153326.10488982286, "data_wait_s": 1.7907999790622853e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 495600, "loss": 0.5824835300445557, "lr": 0.0003, "grad_norm": 0.1449822851167086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72539283354, "tokens_per_second": 150639.1675202296, "data_wait_s": 1.7366000975016505e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ce3e79f2f834d3f2ff1255eaecaa9d1598e5c69d590bb16dcbcfad8000ba500
-size 224121
+oid sha256:589fbaa4f6c29f7af7eb1854870787947a5c6255dd52c74251e33dab1388e512
+size 226239