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": 75586.0442493059,
-      "required_pages_per_second_upper_estimate": 0.00522921792495281
+      "required_token_positions_per_second": 75189.42061106756,
+      "required_pages_per_second_upper_estimate": 0.00520177858136585
     },
     {
       "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": 9448.255531163237,
-      "required_pages_per_second_upper_estimate": 0.003955787288886597
+      "required_token_positions_per_second": 9398.677576383445,
+      "required_pages_per_second_upper_estimate": 0.0039350300345257225
     },
     {
       "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": 18896.511062326474,
-      "required_pages_per_second_upper_estimate": 0.003935644919774346
+      "required_token_positions_per_second": 18797.35515276689,
+      "required_pages_per_second_upper_estimate": 0.003914993358730266
     },
     {
       "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": 28344.76659348971,
-      "required_pages_per_second_upper_estimate": 0.0016992254773271103
+      "required_token_positions_per_second": 28196.032729150334,
+      "required_pages_per_second_upper_estimate": 0.0016903091092634259
     },
     {
       "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": 28344.76659348971,
-      "required_pages_per_second_upper_estimate": 0.002203680927191688
+      "required_token_positions_per_second": 28196.032729150334,
+      "required_pages_per_second_upper_estimate": 0.002192117523450432
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28344.76659348971,
-      "required_pages_per_second_upper_estimate": 0.36357623162208935
+      "required_token_positions_per_second": 28196.032729150334,
+      "required_pages_per_second_upper_estimate": 0.3616684333083251
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08379053826567429,
-    "producer_p95_seconds": 0.2519872410011885,
-    "gpu_update_seconds": 0.9754181573098667,
-    "producer_batches_per_second": 11.934521733579327,
-    "gpu_batches_per_second": 1.0252013380168443,
-    "producer_headroom": 11.641149197742504,
+    "producer_batch_seconds": 0.06661880462493741,
+    "producer_p95_seconds": 0.24485065100088832,
+    "gpu_update_seconds": 0.9805634808834736,
+    "producer_batches_per_second": 15.01077669630941,
+    "gpu_batches_per_second": 1.0198217856318843,
+    "producer_headroom": 14.719019448097683,
     "sustainable": true,
-    "buffer_seconds": 60.47592575321173,
+    "buffer_seconds": 60.79493581477536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387457.7741328725,
-    "upload_bytes_per_second": 4985113.820509082,
+    "generation_bytes_per_second": 387342.0233619356,
+    "upload_bytes_per_second": 4881637.920145985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.13941877355225,
+    "estimated_full_shards_per_hour": 83.11458135264921,
     "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": 1.0211684169989894,
-      "bytes": 10395259,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8835518189989671,
       "bytes": 2669717,
@@ -562,6 +556,12 @@
       "bytes": 10937725,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.8731399669995881,
+      "bytes": 778500,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 418400, "loss": 0.590329110622406, "lr": 0.0003, "grad_norm": 0.17341479540921204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61242727058, "tokens_per_second": 150676.7163422943, "data_wait_s": 1.2142998457420617e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418500, "loss": 0.5561164021492004, "lr": 0.0003, "grad_norm": 0.15359413721278908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61257245218, "tokens_per_second": 150705.3470188801, "data_wait_s": 6.4209998527076095e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 418600, "loss": 0.6041417121887207, "lr": 0.0003, "grad_norm": 0.15793977085570282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61271908979, "tokens_per_second": 152196.6740768357, "data_wait_s": 1.8008999177254736e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 418700, "loss": 0.5904332399368286, "lr": 0.0003, "grad_norm": 0.15509538563603148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61286643721, "tokens_per_second": 152560.1140933915, "data_wait_s": 1.0283001756761223e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eda5d28d82ae12f6a94a7dae5d1a8ae3b2ba4387e99ed0c56142c23329f2dd78
-size 458667
+oid sha256:fdcdcb5c9628fd7cc5fd10ee719a026ed9bb7070fe5e8258a85fc22d99048bb3
+size 460785