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": 75978.97387770619,
-      "required_pages_per_second_upper_estimate": 0.005256401708367891
+      "required_token_positions_per_second": 75996.29245529347,
+      "required_pages_per_second_upper_estimate": 0.005257599847750016
     },
     {
       "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": 9497.371734713273,
-      "required_pages_per_second_upper_estimate": 0.003976351217650035
+      "required_token_positions_per_second": 9499.536556911684,
+      "required_pages_per_second_upper_estimate": 0.003977257583497882
     },
     {
       "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": 18994.743469426547,
-      "required_pages_per_second_upper_estimate": 0.003956104139610509
+      "required_token_positions_per_second": 18999.073113823368,
+      "required_pages_per_second_upper_estimate": 0.003957005890357966
     },
     {
       "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": 28492.11520413982,
-      "required_pages_per_second_upper_estimate": 0.0017080588015472833
+      "required_token_positions_per_second": 28498.60967073505,
+      "required_pages_per_second_upper_estimate": 0.001708448134903191
     },
     {
       "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": 28492.11520413982,
-      "required_pages_per_second_upper_estimate": 0.002215136633551691
+      "required_token_positions_per_second": 28498.60967073505,
+      "required_pages_per_second_upper_estimate": 0.0022156415497633307
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28492.11520413982,
-      "required_pages_per_second_upper_estimate": 0.36546626138889726
+      "required_token_positions_per_second": 28498.60967073505,
+      "required_pages_per_second_upper_estimate": 0.36554956543316597
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09029021917970681,
-    "producer_p95_seconds": 0.2719594899972435,
-    "gpu_update_seconds": 0.9703737262715696,
-    "producer_batches_per_second": 11.07539674933866,
-    "gpu_batches_per_second": 1.0305307871867702,
-    "producer_headroom": 10.747274013591785,
+    "producer_batch_seconds": 0.3949784966406469,
+    "producer_p95_seconds": 0.27786395699877176,
+    "gpu_update_seconds": 0.9701525905802859,
+    "producer_batches_per_second": 2.5317833970840296,
+    "gpu_batches_per_second": 1.0307656854287852,
+    "producer_headroom": 2.4562162214692282,
     "sustainable": true,
-    "buffer_seconds": 60.16317102883731,
+    "buffer_seconds": 60.14946061597772,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441392.01071492716,
-    "upload_bytes_per_second": 4986591.865038698,
+    "generation_bytes_per_second": 448763.14107064257,
+    "upload_bytes_per_second": 5049997.990756401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.71245041929113,
+    "estimated_full_shards_per_hour": 96.29412340249498,
     "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": 1.4380360830000427,
-      "bytes": 7912633,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8056872329998441,
       "bytes": 861106,
@@ -562,6 +556,12 @@
       "bytes": 963365,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.1665040549996775,
+      "bytes": 11979568,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -236,3 +236,4 @@
 {"step": 628000, "loss": 0.5480430126190186, "lr": 0.0001, "grad_norm": 0.1529228731650541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91922483197, "tokens_per_second": 151008.74539826976, "data_wait_s": 1.895099921966903e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 628100, "loss": 0.5606343746185303, "lr": 0.0001, "grad_norm": 0.1497014292572719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91937212305, "tokens_per_second": 152742.78066113414, "data_wait_s": 1.5298999642254785e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 628200, "loss": 0.5492480993270874, "lr": 0.0001, "grad_norm": 0.17231699226995587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91951904881, "tokens_per_second": 151646.3340507062, "data_wait_s": 1.5400997654069215e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 628300, "loss": 0.5330267548561096, "lr": 0.0001, "grad_norm": 0.16012549730276274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91966620376, "tokens_per_second": 147147.75592414188, "data_wait_s": 1.572800101712346e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:def94c13fce8edbe0f955dfde71b701ccb85b33b4633b8f0042ccfb628c77360
-size 508164
+oid sha256:e12faa4a8327902d94a2f8d233c1bb785df5ba0f1d6007ff531aa20483ea8cc7
+size 510282