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": 75719.46777493303,
-      "required_pages_per_second_upper_estimate": 0.005238448474040928
+      "required_token_positions_per_second": 77003.33400925629,
+      "required_pages_per_second_upper_estimate": 0.005327269319111498
     },
     {
       "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": 9464.933471866629,
-      "required_pages_per_second_upper_estimate": 0.003962769994384102
+      "required_token_positions_per_second": 9625.416751157036,
+      "required_pages_per_second_upper_estimate": 0.0040299609921510375
     },
     {
       "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": 18929.866943733257,
-      "required_pages_per_second_upper_estimate": 0.003942592070217633
+      "required_token_positions_per_second": 19250.83350231407,
+      "required_pages_per_second_upper_estimate": 0.004009440939912657
     },
     {
       "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": 28394.800415599886,
-      "required_pages_per_second_upper_estimate": 0.0017022249285653926
+      "required_token_positions_per_second": 28876.250253471106,
+      "required_pages_per_second_upper_estimate": 0.0017310871112144353
     },
     {
       "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": 28394.800415599886,
-      "required_pages_per_second_upper_estimate": 0.002207570836785229
+      "required_token_positions_per_second": 28876.250253471106,
+      "required_pages_per_second_upper_estimate": 0.0022450014440056824
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28394.800415599886,
-      "required_pages_per_second_upper_estimate": 0.3642180117699861
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28876.250253471106,
+      "required_pages_per_second_upper_estimate": 0.3703935333496377
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09042506477311463,
-    "producer_p95_seconds": 0.34070606399473036,
-    "gpu_update_seconds": 0.9736993954995508,
-    "producer_batches_per_second": 11.058880660014989,
-    "gpu_batches_per_second": 1.027011010402195,
-    "producer_headroom": 10.768025413558268,
+    "producer_batch_seconds": 0.11504062230449108,
+    "producer_p95_seconds": 0.335532549994241,
+    "gpu_update_seconds": 0.9574650363987802,
+    "producer_batches_per_second": 8.692581628715345,
+    "gpu_batches_per_second": 1.0444245606724214,
+    "producer_headroom": 8.322842985537307,
     "sustainable": true,
-    "buffer_seconds": 60.369362520972146,
+    "buffer_seconds": 60.320297293123154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405034.97277946403,
-    "upload_bytes_per_second": 4074110.3949593604,
+    "generation_bytes_per_second": 405849.375040509,
+    "upload_bytes_per_second": 4065492.6572068073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.91107642686788,
+    "estimated_full_shards_per_hour": 87.08582819377376,
     "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": 1.8049468739991426,
-      "bytes": 4697774,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 2.496297884001251,
       "bytes": 13693480,
@@ -562,6 +556,12 @@
       "bytes": 11510011,
       "pages": 30,
       "shards": 4
+    },
+    {
+      "seconds": 1.1747664850045112,
+      "bytes": 1216277,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"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}
 {"step": 838100, "loss": 0.5496625304222107, "lr": 3e-05, "grad_norm": 0.3057147170040678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122673957964, "tokens_per_second": 151879.08999256176, "data_wait_s": 1.5837998944334686e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 838200, "loss": 0.5123892426490784, "lr": 3e-05, "grad_norm": 0.29468094144911877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122688655383, "tokens_per_second": 152371.31204812843, "data_wait_s": 8.53499659569934e-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 sha256:3e1938913ce7a1903335cc4b9f0bcd74be8f154431dd4f1915505aa64d883981
-size 812604
+oid sha256:e4b5a1343e22c5cc4da40afda697a0811edd7503e4e1c92be9d8511ce25405e0
+size 814722