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": 76468.12984459836,
-      "required_pages_per_second_upper_estimate": 0.005290242653155707
+      "required_token_positions_per_second": 75863.90080750495,
+      "required_pages_per_second_upper_estimate": 0.00524844068636506
     },
     {
       "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": 9558.516230574794,
-      "required_pages_per_second_upper_estimate": 0.004001951141224986
+      "required_token_positions_per_second": 9482.987600938119,
+      "required_pages_per_second_upper_estimate": 0.003970328881213248
     },
     {
       "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": 19117.03246114959,
-      "required_pages_per_second_upper_estimate": 0.0039815737116088365
+      "required_token_positions_per_second": 18965.975201876237,
+      "required_pages_per_second_upper_estimate": 0.003950112468150013
     },
     {
       "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": 28675.54869172438,
-      "required_pages_per_second_upper_estimate": 0.0017190553590412595
+      "required_token_positions_per_second": 28448.962802814356,
+      "required_pages_per_second_upper_estimate": 0.0017054718809777234
     },
     {
       "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": 28675.54869172438,
-      "required_pages_per_second_upper_estimate": 0.0022293977803727483
+      "required_token_positions_per_second": 28448.962802814356,
+      "required_pages_per_second_upper_estimate": 0.0022117817241559903
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28675.54869172438,
-      "required_pages_per_second_upper_estimate": 0.3678191492120981
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28448.962802814356,
+      "required_pages_per_second_upper_estimate": 0.36491274871813284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09743671146895849,
-    "producer_p95_seconds": 0.36903855000127805,
-    "gpu_update_seconds": 0.9641663808155508,
-    "producer_batches_per_second": 10.263072151389071,
-    "gpu_batches_per_second": 1.0371653896022996,
-    "producer_headroom": 9.89530913225367,
+    "producer_batch_seconds": 0.38359132467934387,
+    "producer_p95_seconds": 0.37708923099853564,
+    "gpu_update_seconds": 0.9718456237450204,
+    "producer_batches_per_second": 2.6069411263039686,
+    "gpu_batches_per_second": 1.028970008782348,
+    "producer_headroom": 2.533544324959426,
     "sustainable": true,
-    "buffer_seconds": 60.742481991379705,
+    "buffer_seconds": 60.25442867219126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430637.05594232515,
-    "upload_bytes_per_second": 4034579.605100602,
+    "generation_bytes_per_second": 396219.91836628545,
+    "upload_bytes_per_second": 4074110.3949593604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.40468748762432,
+    "estimated_full_shards_per_hour": 85.01957095376417,
     "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": 1.5131899659972987,
-      "bytes": 5828306,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.8049468739991426,
       "bytes": 4697774,
@@ -562,6 +556,12 @@
       "bytes": 6148151,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.876006619000691,
+      "bytes": 11510011,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 837700, "loss": 0.5169020295143127, "lr": 3e-05, "grad_norm": 0.28522732600895295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122615260985, "tokens_per_second": 150896.4012463579, "data_wait_s": 1.5616002201568335e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837800, "loss": 0.5466073751449585, "lr": 3e-05, "grad_norm": 0.31192280240627446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122629985080, "tokens_per_second": 152608.73689742765, "data_wait_s": 1.4105993614066392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 837900, "loss": 0.5396285653114319, "lr": 3e-05, "grad_norm": 0.2987354370106703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122644649574, "tokens_per_second": 151569.57592995185, "data_wait_s": 1.748999784467742e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 838000, "loss": 0.5572761297225952, "lr": 3e-05, "grad_norm": 0.2961460656018201, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122659332264, "tokens_per_second": 151520.22936016961, "data_wait_s": 9.369003237225115e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf47421ed1023b99fce13db6f3a6c3a12c358c0eb403acfe2374c827f2513909
-size 808368
+oid sha256:144c0a68b6b3e39ebb04b1d740f67759b068bf56605b3bcda03ddc4e100f2af7
+size 810486