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": 77121.89355917645,
-      "required_pages_per_second_upper_estimate": 0.00533547154387103
+      "required_token_positions_per_second": 77014.38618971394,
+      "required_pages_per_second_upper_estimate": 0.00532803393460015
     },
     {
       "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": 9640.236694897056,
-      "required_pages_per_second_upper_estimate": 0.004036165793119367
+      "required_token_positions_per_second": 9626.798273714243,
+      "required_pages_per_second_upper_estimate": 0.004030539406796268
     },
     {
       "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": 19280.47338979411,
-      "required_pages_per_second_upper_estimate": 0.0040156141468183516
+      "required_token_positions_per_second": 19253.596547428486,
+      "required_pages_per_second_upper_estimate": 0.004010016409343589
     },
     {
       "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": 28920.710084691167,
-      "required_pages_per_second_upper_estimate": 0.0017337524102098464
+      "required_token_positions_per_second": 28880.39482114273,
+      "required_pages_per_second_upper_estimate": 0.0017313355717179608
     },
     {
       "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": 28920.710084691167,
-      "required_pages_per_second_upper_estimate": 0.002248458000324912
+      "required_token_positions_per_second": 28880.39482114273,
+      "required_pages_per_second_upper_estimate": 0.0022453236659121245
     },
     {
       "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": 28920.710084691167,
-      "required_pages_per_second_upper_estimate": 0.37096381632728115
+      "required_token_positions_per_second": 28880.39482114273,
+      "required_pages_per_second_upper_estimate": 0.3704466954136393
     }
   ],
   "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": 0.49352687698441855,
-    "producer_p95_seconds": 0.3021274729981087,
-    "gpu_update_seconds": 0.9559931246167824,
-    "producer_batches_per_second": 2.0262320992734337,
-    "gpu_batches_per_second": 1.0460326274844896,
-    "producer_headroom": 1.9370639557832323,
+    "producer_batch_seconds": 0.10783678921879414,
+    "producer_p95_seconds": 0.2882406610006001,
+    "gpu_update_seconds": 0.9573276325072773,
+    "producer_batches_per_second": 9.273273131037518,
+    "gpu_batches_per_second": 1.044574465463785,
+    "producer_headroom": 8.877560612129493,
     "sustainable": true,
-    "buffer_seconds": 60.22756685085729,
+    "buffer_seconds": 60.311640847958465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286862.73154098884,
-    "upload_bytes_per_second": 3944397.417938787,
+    "generation_bytes_per_second": 288071.341933228,
+    "upload_bytes_per_second": 3913889.079196053,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.554064366075984,
+    "estimated_full_shards_per_hour": 61.81340402124052,
     "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": 1.5038929210022616,
-      "bytes": 12535029,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.2274471419987094,
       "bytes": 6988556,
@@ -562,6 +556,12 @@
       "bytes": 5937086,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 0.9517514979997941,
+      "bytes": 6995458,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 554200, "loss": 0.5900949835777283, "lr": 0.0003, "grad_norm": 0.1890609460630365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81116994918, "tokens_per_second": 152811.06633495176, "data_wait_s": 1.622700074221939e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 554300, "loss": 0.5681555271148682, "lr": 0.0003, "grad_norm": 0.16375250339204897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81131706011, "tokens_per_second": 149020.0752156647, "data_wait_s": 1.206399974762462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 554400, "loss": 0.5891408920288086, "lr": 0.0003, "grad_norm": 0.15826079270071136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81146437215, "tokens_per_second": 153105.11426150508, "data_wait_s": 1.3337001291802153e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 554500, "loss": 0.6127526760101318, "lr": 0.0003, "grad_norm": 0.16399397993627016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81161096845, "tokens_per_second": 153161.7267546149, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db292fd4ee4b4945e480ae807b94fa361938e96693baf7fc4349e716d088abbf
-size 626772
+oid sha256:69079846f388840804ac9d0e0b3c91e9b05aa0d8de2e04b4df9452abd6fe8137
+size 628890