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": 75160.90077883234,
-      "required_pages_per_second_upper_estimate": 0.00651475638594774
+      "required_token_positions_per_second": 75458.95402589241,
+      "required_pages_per_second_upper_estimate": 0.006540590885993859
     },
     {
       "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": 9395.112597354042,
-      "required_pages_per_second_upper_estimate": 0.0037478508845356796
+      "required_token_positions_per_second": 9432.369253236551,
+      "required_pages_per_second_upper_estimate": 0.003762713121603858
     },
     {
       "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": 18790.225194708084,
-      "required_pages_per_second_upper_estimate": 0.004346787902126982
+      "required_token_positions_per_second": 18864.738506473102,
+      "required_pages_per_second_upper_estimate": 0.004364025245414321
     },
     {
       "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": 28185.337792062124,
-      "required_pages_per_second_upper_estimate": 0.0017163391944977281
+      "required_token_positions_per_second": 28297.107759709652,
+      "required_pages_per_second_upper_estimate": 0.0017231453991157627
     },
     {
       "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": 28185.337792062124,
-      "required_pages_per_second_upper_estimate": 0.0021697853252415815
+      "required_token_positions_per_second": 28297.107759709652,
+      "required_pages_per_second_upper_estimate": 0.002178389686750161
     },
     {
       "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": 28185.337792062124,
-      "required_pages_per_second_upper_estimate": 0.3499675651199092
+      "required_token_positions_per_second": 28297.107759709652,
+      "required_pages_per_second_upper_estimate": 0.35135537404807293
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.114082863023242,
-    "producer_p95_seconds": 0.3573294870002428,
-    "gpu_update_seconds": 0.9809355560672594,
-    "producer_batches_per_second": 8.765558415169428,
-    "gpu_batches_per_second": 1.0194349606503952,
-    "producer_headroom": 8.598447918224267,
+    "producer_batch_seconds": 0.12865515918730352,
+    "producer_p95_seconds": 0.34807392400034587,
+    "gpu_update_seconds": 0.9770609856943092,
+    "producer_batches_per_second": 7.772715888868031,
+    "gpu_batches_per_second": 1.0234775665404243,
+    "producer_headroom": 7.594417447899217,
     "sustainable": true,
-    "buffer_seconds": 60.81800447617008,
+    "buffer_seconds": 60.57778111304717,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348709.5658345553,
-    "upload_bytes_per_second": 4147201.545468801,
+    "generation_bytes_per_second": 371741.274192532,
+    "upload_bytes_per_second": 4161727.9075428573,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.8249552848577,
+    "estimated_full_shards_per_hour": 79.76702374774904,
     "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": 0.9854615350013773,
-      "bytes": 5689074,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9606277580023743,
       "bytes": 2676212,
@@ -562,6 +556,12 @@
       "bytes": 5466457,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1624265700011165,
+      "bytes": 7895099,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 345700, "loss": 0.5666700676083565, "lr": 0.0003, "grad_norm": 0.14349329657627782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50601278274, "tokens_per_second": 146473.16475754857, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345800, "loss": 0.5773425437510014, "lr": 0.0003, "grad_norm": 0.1719673929543946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50615996588, "tokens_per_second": 149396.85238823932, "data_wait_s": 1.3353997928788885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345900, "loss": 0.6154396384954453, "lr": 0.0003, "grad_norm": 0.15911986876923162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50630570454, "tokens_per_second": 148408.74728900185, "data_wait_s": 1.6066001990111545e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 346000, "loss": 0.6023015528917313, "lr": 0.0003, "grad_norm": 0.1422033520210749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50645356753, "tokens_per_second": 150134.3683929201, "data_wait_s": 9.304996638093144e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d61dc52fe0ad10a0e47a4defc7490223cb7271abe4e0e0b180f0451b29641d17
-size 659325
+oid sha256:136b33ef4370c6dab06c4132a254f404736a1b31b436382b34de7a93355baf31
+size 661443