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": 76399.13113781046,
-      "required_pages_per_second_upper_estimate": 0.0052854691625210636
+      "required_token_positions_per_second": 76576.98648767339,
+      "required_pages_per_second_upper_estimate": 0.005297773608305849
     },
     {
       "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": 9549.891392226307,
-      "required_pages_per_second_upper_estimate": 0.003998340101515579
+      "required_token_positions_per_second": 9572.123310959174,
+      "required_pages_per_second_upper_estimate": 0.004007648141633776
     },
     {
       "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": 19099.782784452615,
-      "required_pages_per_second_upper_estimate": 0.003977981058857423
+      "required_token_positions_per_second": 19144.246621918348,
+      "required_pages_per_second_upper_estimate": 0.003987241703611287
     },
     {
       "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": 28649.67417667892,
-      "required_pages_per_second_upper_estimate": 0.0017175042213723813
+      "required_token_positions_per_second": 28716.36993287752,
+      "required_pages_per_second_upper_estimate": 0.001721502530117964
     },
     {
       "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": 28649.67417667892,
-      "required_pages_per_second_upper_estimate": 0.0022273861506379275
+      "required_token_positions_per_second": 28716.36993287752,
+      "required_pages_per_second_upper_estimate": 0.0022325714523187407
     },
     {
       "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": 28649.67417667892,
-      "required_pages_per_second_upper_estimate": 0.36748725871498467
+      "required_token_positions_per_second": 28716.36993287752,
+      "required_pages_per_second_upper_estimate": 0.3683427602631767
     }
   ],
   "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": 0.22659291311737206,
-    "producer_p95_seconds": 0.2419307949967333,
-    "gpu_update_seconds": 0.9650371529357813,
-    "producer_batches_per_second": 4.413200687710889,
-    "gpu_batches_per_second": 1.036229534746778,
-    "producer_headroom": 4.2589026270027475,
+    "producer_batch_seconds": 0.45245356269552417,
+    "producer_p95_seconds": 0.2475005529995542,
+    "gpu_update_seconds": 0.962795787372333,
+    "producer_batches_per_second": 2.2101715677569853,
+    "gpu_batches_per_second": 1.0386418523176186,
+    "producer_headroom": 2.12794387480653,
     "sustainable": true,
-    "buffer_seconds": 60.797340634954224,
+    "buffer_seconds": 60.65613460445697,
     "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": 429627.8560105532,
-    "upload_bytes_per_second": 4698084.233941985,
+    "generation_bytes_per_second": 378326.9442568156,
+    "upload_bytes_per_second": 4803917.1719250465,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.1881366752381,
+    "estimated_full_shards_per_hour": 81.18015523699141,
     "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": 1.0195367809992604,
-      "bytes": 6121053,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 2.6576393979994464,
       "bytes": 1839915,
@@ -562,6 +556,12 @@
       "bytes": 2336300,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2755690529993444,
+      "bytes": 16950690,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 782700, "loss": 0.49081888794898987, "lr": 3e-05, "grad_norm": 0.2465137265393661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114564978464, "tokens_per_second": 151430.1650988239, "data_wait_s": 1.4089000615058467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782800, "loss": 0.5230893492698669, "lr": 3e-05, "grad_norm": 0.2622827538015773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114579688632, "tokens_per_second": 152982.13293579072, "data_wait_s": 1.2935000995639712e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 782900, "loss": 0.5246395468711853, "lr": 3e-05, "grad_norm": 0.25183891041606093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114594406316, "tokens_per_second": 152931.21247063638, "data_wait_s": 8.129998604999855e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 783000, "loss": 0.5296126008033752, "lr": 3e-05, "grad_norm": 0.26037759137340166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114609159455, "tokens_per_second": 152681.0293511491, "data_wait_s": 1.2995998986298218e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:792685cf815841fb1e26406a5660a5e8b9204112ee96bb8f224ec57afd81fc54
-size 488319
+oid sha256:980b62fc3dfc19d6e86e43a22f8fe1791f4148b847083c03a7e3aa6bc3f895c0
+size 490437