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": 77086.81215758805,
-      "required_pages_per_second_upper_estimate": 0.005333044531109079
+      "required_token_positions_per_second": 77229.23190869526,
+      "required_pages_per_second_upper_estimate": 0.005342897459950027
     },
     {
       "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": 9635.851519698506,
-      "required_pages_per_second_upper_estimate": 0.0040343298118366054
+      "required_token_positions_per_second": 9653.653988586908,
+      "required_pages_per_second_upper_estimate": 0.004041783333802355
     },
     {
       "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": 19271.703039397013,
-      "required_pages_per_second_upper_estimate": 0.00401378751412033
+      "required_token_positions_per_second": 19307.307977173816,
+      "required_pages_per_second_upper_estimate": 0.004021203083693886
     },
     {
       "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": 28907.55455909552,
-      "required_pages_per_second_upper_estimate": 0.001732963756537712
+      "required_token_positions_per_second": 28960.96196576072,
+      "required_pages_per_second_upper_estimate": 0.0017361654490189026
     },
     {
       "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": 28907.55455909552,
-      "required_pages_per_second_upper_estimate": 0.002247435216074892
+      "required_token_positions_per_second": 28960.96196576072,
+      "required_pages_per_second_upper_estimate": 0.002251587407028756
     },
     {
       "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": 28907.55455909552,
-      "required_pages_per_second_upper_estimate": 0.37079507137024303
+      "required_token_positions_per_second": 28960.96196576072,
+      "required_pages_per_second_upper_estimate": 0.37148012423853877
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1316224329843294,
-    "producer_p95_seconds": 0.25483063400315586,
-    "gpu_update_seconds": 0.9564281870844308,
-    "producer_batches_per_second": 7.597489100653968,
-    "gpu_batches_per_second": 1.0455568055228415,
-    "producer_headroom": 7.266452726932197,
+    "producer_batch_seconds": 0.2828213001405402,
+    "producer_p95_seconds": 0.2611907130012696,
+    "gpu_update_seconds": 0.9546644214611041,
+    "producer_batches_per_second": 3.5358015803727576,
+    "gpu_batches_per_second": 1.0474884970254892,
+    "producer_headroom": 3.375503970127816,
     "sustainable": true,
-    "buffer_seconds": 60.254975786319136,
+    "buffer_seconds": 60.14385855204956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356817.6615964233,
-    "upload_bytes_per_second": 4881853.291856706,
+    "generation_bytes_per_second": 338786.62717738585,
+    "upload_bytes_per_second": 4901889.615367701,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.56476388854527,
+    "estimated_full_shards_per_hour": 72.69572364321881,
     "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": 1.1770325620018411,
-      "bytes": 6029556,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.1039014870002575,
       "bytes": 6886174,
@@ -562,6 +556,12 @@
       "bytes": 936260,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.1434297949999745,
+      "bytes": 7632762,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 784600, "loss": 0.553794801235199, "lr": 3e-05, "grad_norm": 0.2710875305550052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114843234669, "tokens_per_second": 153154.4835286263, "data_wait_s": 7.327998901018873e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 784700, "loss": 0.526593029499054, "lr": 3e-05, "grad_norm": 0.2563428063678736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114857902009, "tokens_per_second": 152240.74937383627, "data_wait_s": 1.8981998437084258e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 784800, "loss": 0.5423645377159119, "lr": 3e-05, "grad_norm": 0.2533305691222203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114872383127, "tokens_per_second": 149908.45026669605, "data_wait_s": 9.030001820065081e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 784900, "loss": 0.5186707973480225, "lr": 3e-05, "grad_norm": 0.25982937823500113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114886902437, "tokens_per_second": 150865.83335429293, "data_wait_s": 1.355299900751561e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a965d6967f12dddf3d97b13d7793e6f2cef55d8e805ba782b9177237a1e54069
-size 528561
+oid sha256:d184b37aa03063cca5d39c21072b07cb61d6076df995fd833492744b356e589c
+size 530679