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": 73869.97188809472,
-      "required_pages_per_second_upper_estimate": 0.006402861941527964
+      "required_token_positions_per_second": 73804.36686174218,
+      "required_pages_per_second_upper_estimate": 0.006397175464118145
     },
     {
       "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": 9233.74648601184,
-      "required_pages_per_second_upper_estimate": 0.00368347953008291
+      "required_token_positions_per_second": 9225.545857717772,
+      "required_pages_per_second_upper_estimate": 0.0036802081768460876
     },
     {
       "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": 18467.49297202368,
-      "required_pages_per_second_upper_estimate": 0.004272129482302072
+      "required_token_positions_per_second": 18451.091715435545,
+      "required_pages_per_second_upper_estimate": 0.004268335340242664
     },
     {
       "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": 27701.239458035518,
-      "required_pages_per_second_upper_estimate": 0.001686860145823175
+      "required_token_positions_per_second": 27676.637573153315,
+      "required_pages_per_second_upper_estimate": 0.0016853620201099635
     },
     {
       "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": 27701.239458035518,
-      "required_pages_per_second_upper_estimate": 0.0021325180954182565
+      "required_token_positions_per_second": 27676.637573153315,
+      "required_pages_per_second_upper_estimate": 0.0021306241742176484
     },
     {
       "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": 27701.239458035518,
-      "required_pages_per_second_upper_estimate": 0.34395668398420004
+      "required_token_positions_per_second": 27676.637573153315,
+      "required_pages_per_second_upper_estimate": 0.3436512109111752
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12461287271892729,
-    "producer_p95_seconds": 0.312866792002751,
-    "gpu_update_seconds": 0.998078083902485,
-    "producer_batches_per_second": 8.024853116544124,
-    "gpu_batches_per_second": 1.0019256169717707,
-    "producer_headroom": 8.009430022159245,
+    "producer_batch_seconds": 0.0747356753517181,
+    "producer_p95_seconds": 0.2817577589994471,
+    "gpu_update_seconds": 0.9989652799015912,
+    "producer_batches_per_second": 13.380490579550386,
+    "gpu_batches_per_second": 1.0010357918530568,
+    "producer_headroom": 13.366645517021155,
     "sustainable": true,
-    "buffer_seconds": 60.882763118051585,
+    "buffer_seconds": 60.93688207399706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436780.0292372014,
-    "upload_bytes_per_second": 4123895.7792860963,
+    "generation_bytes_per_second": 436373.6246805184,
+    "upload_bytes_per_second": 4156526.602018431,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.72282655560522,
+    "estimated_full_shards_per_hour": 93.63562159835494,
     "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": 0.9038558969987207,
-      "bytes": 2334921,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.410665313997015,
       "bytes": 15644643,
@@ -562,6 +556,12 @@
       "bytes": 2678604,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9471035349997692,
+      "bytes": 5896771,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 104400, "loss": 0.5811884552240372, "lr": 0.0003, "grad_norm": 0.1369187457982779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15277309694, "tokens_per_second": 147965.02447973102, "data_wait_s": 1.503100065747276e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104500, "loss": 0.6464366614818573, "lr": 0.0003, "grad_norm": 0.13824862454970732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15292078571, "tokens_per_second": 149769.82605387564, "data_wait_s": 9.871000656858087e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 104600, "loss": 0.6247983071953058, "lr": 0.0003, "grad_norm": 0.13088340708478435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15306817738, "tokens_per_second": 149345.20258404457, "data_wait_s": 1.2896998669020832e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 104700, "loss": 0.5540802646428347, "lr": 0.0003, "grad_norm": 0.13967225520373433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15321526734, "tokens_per_second": 148942.31345991525, "data_wait_s": 1.6287001926684752e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16bdb2d04e81ec8b201ec9b3fda43bd6682705bdcf3b79573873a527d4c70720
-size 417642
+oid sha256:80c18b427c4ae872077b1752a69c231e8be2f8563fcdfef6f2b0c4537682c68c
+size 419760