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": 76699.0809454365,
-      "required_pages_per_second_upper_estimate": 0.0053062203861920735
+      "required_token_positions_per_second": 76575.74870629948,
+      "required_pages_per_second_upper_estimate": 0.005297687975718353
     },
     {
       "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": 9587.385118179562,
-      "required_pages_per_second_upper_estimate": 0.004014037941614176
+      "required_token_positions_per_second": 9571.968588287435,
+      "required_pages_per_second_upper_estimate": 0.004007583362482095
     },
     {
       "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": 19174.770236359123,
-      "required_pages_per_second_upper_estimate": 0.0039935989675374565
+      "required_token_positions_per_second": 19143.93717657487,
+      "required_pages_per_second_upper_estimate": 0.0039871772543068635
     },
     {
       "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": 28762.155354538685,
-      "required_pages_per_second_upper_estimate": 0.00172424729623626
+      "required_token_positions_per_second": 28715.905764862306,
+      "required_pages_per_second_upper_estimate": 0.0017214747039541957
     },
     {
       "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": 28762.155354538685,
-      "required_pages_per_second_upper_estimate": 0.00223613106746411
+      "required_token_positions_per_second": 28715.905764862306,
+      "required_pages_per_second_upper_estimate": 0.0022325353652972173
     },
     {
       "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": 28762.155354538685,
-      "required_pages_per_second_upper_estimate": 0.36893004649169053
+      "required_token_positions_per_second": 28715.905764862306,
+      "required_pages_per_second_upper_estimate": 0.3683368064142623
     }
   ],
   "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": 0.42194949187494046,
-    "producer_p95_seconds": 0.17253562600035366,
-    "gpu_update_seconds": 0.9612631480219416,
-    "producer_batches_per_second": 2.369951900063871,
-    "gpu_batches_per_second": 1.0402978643858032,
-    "producer_headroom": 2.2781474241159785,
+    "producer_batch_seconds": 0.13368790966405442,
+    "producer_p95_seconds": 0.17663147799976286,
+    "gpu_update_seconds": 0.96281135014139,
+    "producer_batches_per_second": 7.480107980691068,
+    "gpu_batches_per_second": 1.0386250638332721,
+    "producer_headroom": 7.201932864092553,
     "sustainable": true,
-    "buffer_seconds": 60.55957832538232,
+    "buffer_seconds": 60.657115058907564,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257863.16097347316,
-    "upload_bytes_per_second": 4248105.89806045,
+    "generation_bytes_per_second": 270291.1797706032,
+    "upload_bytes_per_second": 4763499.2112367535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.33143159773966,
+    "estimated_full_shards_per_hour": 57.998195122132984,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 7683491,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.2777914989992496,
+      "bytes": 16450745,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 725700, "loss": 0.5652655363082886, "lr": 5e-05, "grad_norm": 0.23341700700260484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106222370869, "tokens_per_second": 150301.46178247128, "data_wait_s": 8.197000170184765e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 725800, "loss": 0.5135230422019958, "lr": 5e-05, "grad_norm": 0.24270620585370845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106236911957, "tokens_per_second": 150290.77242181444, "data_wait_s": 1.1297999662929215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725900, "loss": 0.4872758984565735, "lr": 5e-05, "grad_norm": 0.2158895676488853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106251599938, "tokens_per_second": 151354.66117830278, "data_wait_s": 1.0000000656873453e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 726000, "loss": 0.5199756026268005, "lr": 5e-05, "grad_norm": 0.20616051583627748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106266221925, "tokens_per_second": 151293.1910790343, "data_wait_s": 1.1030999303329736e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9f3ec3afa1a9a9ef094a3b5759ddc062b9b8f998ca2f367ec16bbff473d6f45
-size 125910
+oid sha256:7eeec088b0bca7185a39373283d903785c56db389b3f047734ba5f9305a39a72
+size 128028