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": 77017.9599674377,
-      "required_pages_per_second_upper_estimate": 0.0053282811768353856
+      "required_token_positions_per_second": 76576.06998895593,
+      "required_pages_per_second_upper_estimate": 0.005297710202798005
     },
     {
       "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": 9627.244995929712,
-      "required_pages_per_second_upper_estimate": 0.004030726440059265
+      "required_token_positions_per_second": 9572.008748619492,
+      "required_pages_per_second_upper_estimate": 0.004007600176774484
     },
     {
       "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": 19254.489991859424,
-      "required_pages_per_second_upper_estimate": 0.0040102024902568395
+      "required_token_positions_per_second": 19144.017497238983,
+      "required_pages_per_second_upper_estimate": 0.003987193982983
     },
     {
       "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": 28881.73498778914,
-      "required_pages_per_second_upper_estimate": 0.001731415912662097
+      "required_token_positions_per_second": 28716.026245858473,
+      "required_pages_per_second_upper_estimate": 0.001721481926605952
     },
     {
       "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": 28881.73498778914,
-      "required_pages_per_second_upper_estimate": 0.0022454278579741127
+      "required_token_positions_per_second": 28716.026245858473,
+      "required_pages_per_second_upper_estimate": 0.0022325447321647294
     },
     {
       "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": 28881.73498778914,
-      "required_pages_per_second_upper_estimate": 0.37046388563242055
+      "required_token_positions_per_second": 28716.026245858473,
+      "required_pages_per_second_upper_estimate": 0.36833835181511876
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11237941836719756,
-    "producer_p95_seconds": 0.3184784969998873,
-    "gpu_update_seconds": 0.9572832107104803,
-    "producer_batches_per_second": 8.89842654935728,
-    "gpu_batches_per_second": 1.0446229379264012,
-    "producer_headroom": 8.518314337440119,
+    "producer_batch_seconds": 0.08890272795298415,
+    "producer_p95_seconds": 0.31213750700408127,
+    "gpu_update_seconds": 0.9628073105688671,
+    "producer_batches_per_second": 11.248248766099124,
+    "gpu_batches_per_second": 1.0386294215081913,
+    "producer_headroom": 10.829896143097475,
     "sustainable": true,
-    "buffer_seconds": 60.30884227476026,
+    "buffer_seconds": 60.65686056583863,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289159.93489250704,
-    "upload_bytes_per_second": 3600010.0695605013,
+    "generation_bytes_per_second": 312903.6359664805,
+    "upload_bytes_per_second": 3624552.0527698924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.046990729154665,
+    "estimated_full_shards_per_hour": 67.14183625455676,
     "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": 1.0961665050017473,
-      "bytes": 2428850,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 6.228057981999882,
       "bytes": 11900976,
@@ -562,6 +556,12 @@
       "bytes": 6796671,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.7764992299998994,
+      "bytes": 7789659,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 882000, "loss": 0.5371716022491455, "lr": 3e-05, "grad_norm": 0.31736149239073724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129099909604, "tokens_per_second": 151683.04700585108, "data_wait_s": 1.6454003343824297e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882100, "loss": 0.5230187773704529, "lr": 3e-05, "grad_norm": 0.31251775305386836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129114604171, "tokens_per_second": 152962.4350397474, "data_wait_s": 1.761299790814519e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882200, "loss": 0.5362783074378967, "lr": 3e-05, "grad_norm": 0.29593270333345456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129129319154, "tokens_per_second": 152992.74771197283, "data_wait_s": 1.4457000361289829e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 882300, "loss": 0.5350361466407776, "lr": 3e-05, "grad_norm": 0.3114949750423096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129144052720, "tokens_per_second": 152699.12172993395, "data_wait_s": 1.0168005246669054e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d7eaed145680b4007d6dfa2f9ad91c3f0653968622f58a8bb81ed492869a47b
-size 847275
+oid sha256:00bcba431eb0cec900ff2beb766ad40985e77449e644108fb744facb5dc97a1a
+size 849393