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": 77319.45146869124,
-      "required_pages_per_second_upper_estimate": 0.005349139058448769
+      "required_token_positions_per_second": 76879.8751427326,
+      "required_pages_per_second_upper_estimate": 0.005318728148261352
     },
     {
       "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": 9664.931433586406,
-      "required_pages_per_second_upper_estimate": 0.0040465049645238135
+      "required_token_positions_per_second": 9609.984392841576,
+      "required_pages_per_second_upper_estimate": 0.004023499785988642
     },
     {
       "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": 19329.86286717281,
-      "required_pages_per_second_upper_estimate": 0.0040259006724682285
+      "required_token_positions_per_second": 19219.96878568315,
+      "required_pages_per_second_upper_estimate": 0.00400301263339575
     },
     {
       "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": 28994.794300759215,
-      "required_pages_per_second_upper_estimate": 0.0017381936458430779
+      "required_token_positions_per_second": 28829.953178524727,
+      "required_pages_per_second_upper_estimate": 0.0017283116722629401
     },
     {
       "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": 28994.794300759215,
-      "required_pages_per_second_upper_estimate": 0.0022542177222621743
+      "required_token_positions_per_second": 28829.953178524727,
+      "required_pages_per_second_upper_estimate": 0.002241402050067913
     },
     {
       "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": 28994.794300759215,
-      "required_pages_per_second_upper_estimate": 0.37191408910556834
+      "required_token_positions_per_second": 28829.953178524727,
+      "required_pages_per_second_upper_estimate": 0.3697996841821517
     }
   ],
   "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": 0.13783576233583972,
-    "producer_p95_seconds": 0.33346579399949405,
-    "gpu_update_seconds": 0.9535504792070917,
-    "producer_batches_per_second": 7.255011203576319,
-    "gpu_batches_per_second": 1.048712178123525,
-    "producer_headroom": 6.918019409823018,
+    "producer_batch_seconds": 0.18355684407060835,
+    "producer_p95_seconds": 0.32356694500049343,
+    "gpu_update_seconds": 0.9590025980546802,
+    "producer_batches_per_second": 5.447903645670289,
+    "gpu_batches_per_second": 1.042750042626039,
+    "producer_headroom": 5.224553750149371,
     "sustainable": true,
-    "buffer_seconds": 60.07368019004677,
+    "buffer_seconds": 60.417163677444854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391901.1784904546,
-    "upload_bytes_per_second": 5251550.401724118,
+    "generation_bytes_per_second": 392753.86398247967,
+    "upload_bytes_per_second": 5235556.367065918,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.09286991152982,
+    "estimated_full_shards_per_hour": 84.27583636861603,
     "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": 1.1000836000021081,
-      "bytes": 6163961,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1141700879998098,
       "bytes": 11677393,
@@ -562,6 +556,12 @@
       "bytes": 13753578,
       "pages": 49,
       "shards": 4
+    },
+    {
+      "seconds": 1.2824578229992767,
+      "bytes": 5780729,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 638000, "loss": 0.5121858716011047, "lr": 0.0001, "grad_norm": 0.18508731141413415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93386675116, "tokens_per_second": 143104.28380744057, "data_wait_s": 1.0420000762678683e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638100, "loss": 0.5641270279884338, "lr": 0.0001, "grad_norm": 0.20221628744098225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93401303482, "tokens_per_second": 152214.43562992098, "data_wait_s": 1.549000444356352e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638200, "loss": 0.5692286491394043, "lr": 0.0001, "grad_norm": 0.1775209778022932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93416081425, "tokens_per_second": 153340.83732563278, "data_wait_s": 9.074996341951191e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 638300, "loss": 0.6117377877235413, "lr": 0.0001, "grad_norm": 0.21144039233800835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93430705861, "tokens_per_second": 151678.49098972362, "data_wait_s": 1.4162003935780376e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efbcea4d07f4f37ef20ea487d75e208fdd747b7d4caf88a4e76765a310e54382
-size 721530
+oid sha256:88d131633a2628bc08076b50f1c28e5ce975beab651f4d4985383aa88a12ef42
+size 723648