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": 77083.87575995187,
-      "required_pages_per_second_upper_estimate": 0.005332841384307227
+      "required_token_positions_per_second": 77142.4691391325,
+      "required_pages_per_second_upper_estimate": 0.005336895010234314
     },
     {
       "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": 9635.484469993984,
-      "required_pages_per_second_upper_estimate": 0.004034176135790191
+      "required_token_positions_per_second": 9642.808642391563,
+      "required_pages_per_second_upper_estimate": 0.004037242613826946
     },
     {
       "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": 19270.968939987968,
-      "required_pages_per_second_upper_estimate": 0.004013634620572929
+      "required_token_positions_per_second": 19285.617284783126,
+      "required_pages_per_second_upper_estimate": 0.004016685484490931
     },
     {
       "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": 28906.453409981954,
-      "required_pages_per_second_upper_estimate": 0.0017328977443297109
+      "required_token_positions_per_second": 28928.42592717469,
+      "required_pages_per_second_upper_estimate": 0.0017342149631853242
     },
     {
       "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": 28906.453409981954,
-      "required_pages_per_second_upper_estimate": 0.0022473496065747553
+      "required_token_positions_per_second": 28928.42592717469,
+      "required_pages_per_second_upper_estimate": 0.0022490578731396015
     },
     {
       "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": 28906.453409981954,
-      "required_pages_per_second_upper_estimate": 0.37078094701173603
+      "required_token_positions_per_second": 28928.42592717469,
+      "required_pages_per_second_upper_estimate": 0.3710627868700336
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11050502215573488,
-    "producer_p95_seconds": 0.3499120169944945,
-    "gpu_update_seconds": 0.9564646208189834,
-    "producer_batches_per_second": 9.049362467804393,
-    "gpu_batches_per_second": 1.0455169780809446,
-    "producer_headroom": 8.655395041422068,
+    "producer_batch_seconds": 0.38847792917198376,
+    "producer_p95_seconds": 0.33627909300412284,
+    "gpu_update_seconds": 0.9557381403883477,
+    "producer_batches_per_second": 2.57414881234421,
+    "gpu_batches_per_second": 1.04631170164839,
+    "producer_headroom": 2.460212198992729,
     "sustainable": true,
-    "buffer_seconds": 60.257271111595955,
+    "buffer_seconds": 60.211502844465905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414997.90678260865,
-    "upload_bytes_per_second": 5808655.167919922,
+    "generation_bytes_per_second": 395281.8314537348,
+    "upload_bytes_per_second": 5706225.098094306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.04889013870901,
+    "estimated_full_shards_per_hour": 84.81827933987648,
     "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": [
-    {
-      "seconds": 1.25598544499735,
-      "bytes": 11955544,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0177433429998928,
       "bytes": 6488753,
@@ -562,6 +556,12 @@
       "bytes": 5973673,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9036034050004673,
+      "bytes": 2269487,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -341,3 +341,4 @@
 {"step": 519100, "loss": 0.6098043322563171, "lr": 0.0003, "grad_norm": 0.16214481521855956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75978985534, "tokens_per_second": 155114.9035491838, "data_wait_s": 1.7710000975057483e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519200, "loss": 0.5777155160903931, "lr": 0.0003, "grad_norm": 0.16296838538223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75993615308, "tokens_per_second": 152380.59228125404, "data_wait_s": 1.545400300528854e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519300, "loss": 0.586578905582428, "lr": 0.0003, "grad_norm": 0.14709388063529844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76008114939, "tokens_per_second": 150946.43269616063, "data_wait_s": 2.08690034924075e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 519400, "loss": 0.5922352075576782, "lr": 0.0003, "grad_norm": 0.14604102382617246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76022613171, "tokens_per_second": 151098.03143141948, "data_wait_s": 1.3692995707970113e-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:e9777a6e03875d1720f227631729076c460e950badc381f0a6b84d788246e089
-size 731337
+oid sha256:429f7bf0eac5af2a0deb3d4b680f84a291fd3eae237774de8723e4e078e08e8c
+size 733455