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": 75406.63110977021,
-      "required_pages_per_second_upper_estimate": 0.006536055668235605
+      "required_token_positions_per_second": 75115.19973163799,
+      "required_pages_per_second_upper_estimate": 0.006510795135005179
     },
     {
       "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": 9425.828888721277,
-      "required_pages_per_second_upper_estimate": 0.0037601040724115514
+      "required_token_positions_per_second": 9389.399966454748,
+      "required_pages_per_second_upper_estimate": 0.0037455720306585082
     },
     {
       "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": 18851.657777442553,
-      "required_pages_per_second_upper_estimate": 0.0043609992489660725
+      "required_token_positions_per_second": 18778.799932909496,
+      "required_pages_per_second_upper_estimate": 0.004344144868887618
     },
     {
       "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": 28277.48666616383,
-      "required_pages_per_second_upper_estimate": 0.001721950577462744
+      "required_token_positions_per_second": 28168.199899364245,
+      "required_pages_per_second_upper_estimate": 0.0017152955867480012
     },
     {
       "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": 28277.48666616383,
-      "required_pages_per_second_upper_estimate": 0.0021768792006543404
+      "required_token_positions_per_second": 28168.199899364245,
+      "required_pages_per_second_upper_estimate": 0.002168466002820975
     },
     {
       "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": 28277.48666616383,
-      "required_pages_per_second_upper_estimate": 0.35111174573380965
+      "required_token_positions_per_second": 28168.199899364245,
+      "required_pages_per_second_upper_estimate": 0.3497547698494387
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11800769737487826,
-    "producer_p95_seconds": 0.2592683959992428,
-    "gpu_update_seconds": 0.9777389456992633,
-    "producer_batches_per_second": 8.47402349376645,
-    "gpu_batches_per_second": 1.022767891571319,
-    "producer_headroom": 8.285382796626,
+    "producer_batch_seconds": 0.4214766347343897,
+    "producer_p95_seconds": 0.31196988800002146,
+    "gpu_update_seconds": 0.9815323697920794,
+    "producer_batches_per_second": 2.3726107631807154,
+    "gpu_batches_per_second": 1.018815100526774,
+    "producer_headroom": 2.3287942649789617,
     "sustainable": true,
-    "buffer_seconds": 60.61981463335432,
+    "buffer_seconds": 60.85500692710892,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285357.0789660668,
-    "upload_bytes_per_second": 5902045.0231838,
+    "generation_bytes_per_second": 264133.08951264166,
+    "upload_bytes_per_second": 5871870.940823608,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.23098637329582,
+    "estimated_full_shards_per_hour": 56.67681230577886,
     "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": 0.9827494309993199,
-      "bytes": 7704375,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9620766659991205,
       "bytes": 6846709,
@@ -562,6 +556,12 @@
       "bytes": 6977422,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3117401380004594,
+      "bytes": 7488136,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -156,3 +156,4 @@
 {"step": 130600, "loss": 0.5643460601568222, "lr": 0.0003, "grad_norm": 0.13814585154836584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19112736787, "tokens_per_second": 150510.5215499491, "data_wait_s": 1.17090003186604e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 130700, "loss": 0.6164646344259381, "lr": 0.0003, "grad_norm": 0.14307007665930865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19127539470, "tokens_per_second": 150657.64392787468, "data_wait_s": 1.5667999832658097e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 130800, "loss": 0.614993249066174, "lr": 0.0003, "grad_norm": 0.13111472899018056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19142127551, "tokens_per_second": 148461.5193809926, "data_wait_s": 1.2720000086119398e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 130900, "loss": 0.627563763409853, "lr": 0.0003, "grad_norm": 0.13571884190445932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19156667235, "tokens_per_second": 148638.82289956714, "data_wait_s": 1.6165999113582075e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71aff36b38ddde3d2a566701b4af037cc0913749fb82ef4e54a3ed02d174ba21
-size 337158
+oid sha256:65cfe97ac1ef5f7b0f3858f92543e3cbbc2ea3e89ce28dfcbc7da7ec2b53804b
+size 339276