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": 74376.00274689774,
-      "required_pages_per_second_upper_estimate": 0.0064467234138455235
+      "required_token_positions_per_second": 73744.55007475404,
+      "required_pages_per_second_upper_estimate": 0.006391990696626273
     },
     {
       "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": 9297.000343362217,
-      "required_pages_per_second_upper_estimate": 0.0037087124395094213
+      "required_token_positions_per_second": 9218.068759344254,
+      "required_pages_per_second_upper_estimate": 0.003677225450512308
     },
     {
       "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": 18594.000686724434,
-      "required_pages_per_second_upper_estimate": 0.0043013948156383555
+      "required_token_positions_per_second": 18436.13751868851,
+      "required_pages_per_second_upper_estimate": 0.004264875950010111
     },
     {
       "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": 27891.00103008665,
-      "required_pages_per_second_upper_estimate": 0.0016984156299590647
+      "required_token_positions_per_second": 27654.206278032765,
+      "required_pages_per_second_upper_estimate": 0.001683996071925032
     },
     {
       "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": 27891.00103008665,
-      "required_pages_per_second_upper_estimate": 0.002147126466528404
+      "required_token_positions_per_second": 27654.206278032765,
+      "required_pages_per_second_upper_estimate": 0.0021288973510254694
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27891.00103008665,
-      "required_pages_per_second_upper_estimate": 0.34631288761794765
+      "required_token_positions_per_second": 27654.206278032765,
+      "required_pages_per_second_upper_estimate": 0.3433726892984934
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07907547732028775,
+    "producer_batch_seconds": 0.11335760078910084,
     "producer_p95_seconds": 0.21670200799962913,
-    "gpu_update_seconds": 0.9912874754898715,
-    "producer_batches_per_second": 12.646145605287902,
-    "gpu_batches_per_second": 1.008789099757185,
-    "producer_headroom": 12.535965751743177,
+    "gpu_update_seconds": 0.9997755756223713,
+    "producer_batches_per_second": 8.821640481439587,
+    "gpu_batches_per_second": 1.000224474755236,
+    "producer_headroom": 8.819660690264875,
     "sustainable": true,
-    "buffer_seconds": 60.46853600488216,
+    "buffer_seconds": 60.98631011296465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326785.19686125865,
-    "upload_bytes_per_second": 5321500.534583574,
+    "generation_bytes_per_second": 313233.3398521963,
+    "upload_bytes_per_second": 5315232.588522313,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.12049607637711,
+    "estimated_full_shards_per_hour": 67.21258303331773,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 7422959,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2087593340002059,
+      "bytes": 6148456,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 202200, "loss": 0.5846894299611449, "lr": 0.0003, "grad_norm": 0.14067985833372987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29594124743, "tokens_per_second": 148035.92941285914, "data_wait_s": 2.010000025620684e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202300, "loss": 0.6223247908055782, "lr": 0.0003, "grad_norm": 0.1508217402394871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29608864297, "tokens_per_second": 148892.05627461962, "data_wait_s": 2.1283999558363575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202400, "loss": 0.6051944028586149, "lr": 0.0003, "grad_norm": 0.15517594246542166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29623586994, "tokens_per_second": 148461.85263258754, "data_wait_s": 2.168100036215037e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 202500, "loss": 0.6498415544629097, "lr": 0.0003, "grad_norm": 0.15196157946285538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29638299561, "tokens_per_second": 148997.07727475307, "data_wait_s": 2.0152000615780707e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0a14d47f8cdff821506978b30cb1e3e36d3a391eadf7d8985ad461d93b7c4fb
-size 157680
+oid sha256:2129a3a5d738e9b69111f6b7110c8aeeaeb39b1a6e0ecd5edfc45c966cbf887d
+size 159798