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": 76550.43485735178,
-      "required_pages_per_second_upper_estimate": 0.005295936704911929
+      "required_token_positions_per_second": 76341.54110939047,
+      "required_pages_per_second_upper_estimate": 0.005281484950727692
     },
     {
       "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": 9568.804357168972,
-      "required_pages_per_second_upper_estimate": 0.004006258564989454
+      "required_token_positions_per_second": 9542.692638673809,
+      "required_pages_per_second_upper_estimate": 0.003995326133730217
     },
     {
       "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": 19137.608714337945,
-      "required_pages_per_second_upper_estimate": 0.003985859202515684
+      "required_token_positions_per_second": 19085.385277347617,
+      "required_pages_per_second_upper_estimate": 0.003974982437815247
     },
     {
       "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": 28706.41307150692,
-      "required_pages_per_second_upper_estimate": 0.0017209056314820437
+      "required_token_positions_per_second": 28628.07791602143,
+      "required_pages_per_second_upper_estimate": 0.0017162095585215454
     },
     {
       "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": 28706.41307150692,
-      "required_pages_per_second_upper_estimate": 0.0022317973501427813
+      "required_token_positions_per_second": 28628.07791602143,
+      "required_pages_per_second_upper_estimate": 0.0022257071363637193
     },
     {
       "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": 28706.41307150692,
-      "required_pages_per_second_upper_estimate": 0.36821504433635943
+      "required_token_positions_per_second": 28628.07791602143,
+      "required_pages_per_second_upper_estimate": 0.36721024507152844
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2695831617967599,
-    "producer_p95_seconds": 0.24042049499985296,
-    "gpu_update_seconds": 0.9631297344997287,
-    "producer_batches_per_second": 3.7094304901502158,
-    "gpu_batches_per_second": 1.0382817227830916,
-    "producer_headroom": 3.572662803123576,
+    "producer_batch_seconds": 0.06328312953122861,
+    "producer_p95_seconds": 0.2050158799993369,
+    "gpu_update_seconds": 0.9657651513001878,
+    "producer_batches_per_second": 15.8019997969052,
+    "gpu_batches_per_second": 1.035448419994988,
+    "producer_headroom": 15.261020724703688,
     "sustainable": true,
-    "buffer_seconds": 60.67717327348291,
+    "buffer_seconds": 60.84320453191183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 506639.92217322264,
-    "upload_bytes_per_second": 5145170.27868784,
+    "generation_bytes_per_second": 506232.8447274945,
+    "upload_bytes_per_second": 5123872.261933244,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.71313332459935,
+    "estimated_full_shards_per_hour": 108.62578398102404,
     "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": 0.9096585289998984,
-      "bytes": 2702720,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.212294140999802,
       "bytes": 5935208,
@@ -562,6 +556,12 @@
       "bytes": 11552086,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.7776427819990204,
+      "bytes": 236866,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -146,3 +146,4 @@
 {"step": 454800, "loss": 0.5946531891822815, "lr": 0.0003, "grad_norm": 0.15826047422264783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66568790965, "tokens_per_second": 152278.17276115512, "data_wait_s": 1.2017000699415803e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454900, "loss": 0.5882903337478638, "lr": 0.0003, "grad_norm": 0.18120452433200254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66583435758, "tokens_per_second": 152681.19558098027, "data_wait_s": 9.585999578121118e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 455000, "loss": 0.5725584626197815, "lr": 0.0003, "grad_norm": 0.14253844025443568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66598128293, "tokens_per_second": 152960.83429289557, "data_wait_s": 1.4156999895931222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 455100, "loss": 0.6503855586051941, "lr": 0.0003, "grad_norm": 0.25396449083049755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66612713585, "tokens_per_second": 128559.36848418156, "data_wait_s": 1.1336000170558691e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae49be1f8ebf28ae0a4f76dca10b14d763812fa66860b6a609808adeca24bdf2
-size 315978
+oid sha256:8040ccb8db1074bdfa325c776ac67e3b84acc667762d2bc266aa8b6b8672465f
+size 318096