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": 77312.5611724073,
-      "required_pages_per_second_upper_estimate": 0.005348662371764156
+      "required_token_positions_per_second": 77347.42203138491,
+      "required_pages_per_second_upper_estimate": 0.005351074126876564
     },
     {
       "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": 9664.070146550912,
-      "required_pages_per_second_upper_estimate": 0.004046144361627771
+      "required_token_positions_per_second": 9668.427753923113,
+      "required_pages_per_second_upper_estimate": 0.004047968800837322
     },
     {
       "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": 19328.140293101824,
-      "required_pages_per_second_upper_estimate": 0.004025541905716574
+      "required_token_positions_per_second": 19336.855507846227,
+      "required_pages_per_second_upper_estimate": 0.004027357055112165
     },
     {
       "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": 28992.210439652736,
-      "required_pages_per_second_upper_estimate": 0.001738038747315072
+      "required_token_positions_per_second": 29005.283261769342,
+      "required_pages_per_second_upper_estimate": 0.0017388224430399189
     },
     {
       "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": 28992.210439652736,
-      "required_pages_per_second_upper_estimate": 0.0022540168384263494
+      "required_token_positions_per_second": 29005.283261769342,
+      "required_pages_per_second_upper_estimate": 0.002255033192844648
     },
     {
       "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": 28992.210439652736,
-      "required_pages_per_second_upper_estimate": 0.37188094610962835
+      "required_token_positions_per_second": 29005.283261769342,
+      "required_pages_per_second_upper_estimate": 0.3720486302352374
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07969858355465931,
-    "producer_p95_seconds": 0.17926300299950526,
-    "gpu_update_seconds": 0.9536354621028049,
-    "producer_batches_per_second": 12.547274435739183,
-    "gpu_batches_per_second": 1.0486187224990138,
-    "producer_headroom": 11.965525854656846,
+    "producer_batch_seconds": 0.32379250135934967,
+    "producer_p95_seconds": 0.1967074499989394,
+    "gpu_update_seconds": 0.9532056539658649,
+    "producer_batches_per_second": 3.0883976491172205,
+    "gpu_batches_per_second": 1.0490915531600602,
+    "producer_headroom": 2.94387810083342,
     "sustainable": true,
-    "buffer_seconds": 60.07903411247671,
+    "buffer_seconds": 60.051956199849485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375799.296033831,
-    "upload_bytes_per_second": 3072757.0378513853,
+    "generation_bytes_per_second": 355643.867552128,
+    "upload_bytes_per_second": 3116815.615778697,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.6377807689781,
+    "estimated_full_shards_per_hour": 76.31289501116638,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 6243705,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0855320710015803,
+      "bytes": 8387056,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 695600, "loss": 0.5088210701942444, "lr": 8e-05, "grad_norm": 0.21322358441997738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101817276722, "tokens_per_second": 153254.87645242005, "data_wait_s": 1.4193999959388748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695700, "loss": 0.525488018989563, "lr": 8e-05, "grad_norm": 0.19512356613865714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101831984681, "tokens_per_second": 153847.54823968897, "data_wait_s": 1.5897998309810646e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695800, "loss": 0.5793548822402954, "lr": 8e-05, "grad_norm": 0.21585714774418002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101846698572, "tokens_per_second": 153571.48304085416, "data_wait_s": 1.3446000593830831e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 695900, "loss": 0.5242658257484436, "lr": 8e-05, "grad_norm": 0.20053522714558836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101861235743, "tokens_per_second": 151289.95061557947, "data_wait_s": 1.521600097476039e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d73c39dff709f16f949c0d36cdf626794818691de4fed4cf48c65d1edf21c95
-size 230475
+oid sha256:b20ee40df8cbebbf4c0f67f22e23be8f439bef494bb5fe34a1f3d05bcb3fa6c1
+size 232593