Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75688.26221915733,
-      "required_pages_per_second_upper_estimate": 0.00656046673900976
+      "required_token_positions_per_second": 75514.87996795733,
+      "required_pages_per_second_upper_estimate": 0.006545438404909041
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9461.032777394666,
-      "required_pages_per_second_upper_estimate": 0.003774147429948407
+      "required_token_positions_per_second": 9439.359995994666,
+      "required_pages_per_second_upper_estimate": 0.0037655018334109883
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18922.065554789333,
-      "required_pages_per_second_upper_estimate": 0.00437728684912069
+      "required_token_positions_per_second": 18878.719991989332,
+      "required_pages_per_second_upper_estimate": 0.004367259616022761
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28383.098332184,
-      "required_pages_per_second_upper_estimate": 0.001728381773821784
+      "required_token_positions_per_second": 28318.079987983998,
+      "required_pages_per_second_upper_estimate": 0.0017244224977848923
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28383.098332184,
-      "required_pages_per_second_upper_estimate": 0.002185009479056376
+      "required_token_positions_per_second": 28318.079987983998,
+      "required_pages_per_second_upper_estimate": 0.0021800041869375637
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28383.098332184,
-      "required_pages_per_second_upper_estimate": 0.3524230891662714
+      "required_token_positions_per_second": 28318.079987983998,
+      "required_pages_per_second_upper_estimate": 0.35161577893370743
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11359239407019572,
-    "producer_p95_seconds": 0.2926016389974393,
-    "gpu_update_seconds": 0.9741008425655046,
-    "producer_batches_per_second": 8.803406321219347,
-    "gpu_batches_per_second": 1.026587757963831,
-    "producer_headroom": 8.575405514946254,
+    "producer_batch_seconds": 0.12827800925779798,
+    "producer_p95_seconds": 0.32516657099768054,
+    "gpu_update_seconds": 0.9763373792196248,
+    "producer_batches_per_second": 7.795568435976569,
+    "gpu_batches_per_second": 1.0242361106765046,
+    "producer_headroom": 7.611104856308593,
     "sustainable": true,
-    "buffer_seconds": 60.39425223906128,
+    "buffer_seconds": 60.53291751161674,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310024.856623506,
+    "generation_bytes_per_second": 379891.0235785658,
     "upload_bytes_per_second": 4921676.045441605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.52411722210775,
+    "estimated_full_shards_per_hour": 81.51577024953585,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 268000, "loss": 0.5497099291533232, "lr": 0.0003, "grad_norm": 0.14825696052239232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39227199752, "tokens_per_second": 149621.39817186145, "data_wait_s": 1.5390000044135377e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268100, "loss": 0.5864941235631704, "lr": 0.0003, "grad_norm": 0.14459926591310038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39241925666, "tokens_per_second": 149693.1115663491, "data_wait_s": 1.1463998816907406e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268200, "loss": 0.540740785188973, "lr": 0.0003, "grad_norm": 0.14683682745459853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39256643236, "tokens_per_second": 149615.4632257869, "data_wait_s": 1.343500116490759e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 268300, "loss": 0.6001504268497229, "lr": 0.0003, "grad_norm": 0.16388218285261316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39271157663, "tokens_per_second": 148489.45665295143, "data_wait_s": 1.3158998626749963e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0029e46358164b829331ffd446c49682e2888394fa06cdf4139494815c4998f7
-size 663561
+oid sha256:a4e4e59a26cb4277edb8ab381dc45cb4ead810ca429bd7973e232a4a4ee18313
+size 665679