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": 76693.72142881632,
-      "required_pages_per_second_upper_estimate": 0.005305849602396498
+      "required_token_positions_per_second": 77364.78920939543,
+      "required_pages_per_second_upper_estimate": 0.005352275628548737
     },
     {
       "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": 9586.71517860204,
-      "required_pages_per_second_upper_estimate": 0.004013757451902997
+      "required_token_positions_per_second": 9670.598651174429,
+      "required_pages_per_second_upper_estimate": 0.004048877710182958
     },
     {
       "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": 19173.43035720408,
-      "required_pages_per_second_upper_estimate": 0.003993319906044451
+      "required_token_positions_per_second": 19341.197302348857,
+      "required_pages_per_second_upper_estimate": 0.004028261336406242
     },
     {
       "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": 28760.145535806125,
-      "required_pages_per_second_upper_estimate": 0.0017241268106720586
+      "required_token_positions_per_second": 29011.795953523288,
+      "required_pages_per_second_upper_estimate": 0.0017392128689662633
     },
     {
       "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": 28760.145535806125,
-      "required_pages_per_second_upper_estimate": 0.0022359748128979117
+      "required_token_positions_per_second": 29011.795953523288,
+      "required_pages_per_second_upper_estimate": 0.0022555395259822135
     },
     {
       "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": 28760.145535806125,
-      "required_pages_per_second_upper_estimate": 0.36890426669496446
+      "required_token_positions_per_second": 29011.795953523288,
+      "required_pages_per_second_upper_estimate": 0.3721321680522734
     }
   ],
   "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": 0.11999967912498732,
-    "producer_p95_seconds": 0.35848651700507617,
-    "gpu_update_seconds": 0.9613303230881947,
-    "producer_batches_per_second": 8.333355616379908,
-    "gpu_batches_per_second": 1.040225171289284,
-    "producer_headroom": 8.011107447103319,
+    "producer_batch_seconds": 0.14708607803140694,
+    "producer_p95_seconds": 0.3600123200012604,
+    "gpu_update_seconds": 0.952991674293171,
+    "producer_batches_per_second": 6.798739985347032,
+    "gpu_batches_per_second": 1.049327110587503,
+    "producer_headroom": 6.479142601719797,
     "sustainable": true,
-    "buffer_seconds": 60.56381035455627,
+    "buffer_seconds": 60.03847548046977,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296665.1559829677,
-    "upload_bytes_per_second": 4824170.536422267,
+    "generation_bytes_per_second": 285641.51924237894,
+    "upload_bytes_per_second": 4857292.906146977,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.65743646256231,
+    "estimated_full_shards_per_hour": 61.29202063516165,
     "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": 1.4946348649973515,
-      "bytes": 6250315,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1129696229982073,
       "bytes": 1097015,
@@ -562,6 +556,12 @@
       "bytes": 5555010,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1725506970033166,
+      "bytes": 7635277,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 596000, "loss": 0.5765054821968079, "lr": 0.0003, "grad_norm": 0.15422873717060467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87236448609, "tokens_per_second": 150704.33241506864, "data_wait_s": 1.3217999367043376e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596100, "loss": 0.6018339991569519, "lr": 0.0003, "grad_norm": 0.17259046529465089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87251014095, "tokens_per_second": 137373.049811316, "data_wait_s": 1.0986004781443626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 596200, "loss": 0.5732657313346863, "lr": 0.0003, "grad_norm": 0.18199784911604647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87265546722, "tokens_per_second": 150874.52269238717, "data_wait_s": 1.0335003025829792e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 596300, "loss": 0.5600859522819519, "lr": 0.0003, "grad_norm": 0.1622003101067814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87280233172, "tokens_per_second": 153579.91889956864, "data_wait_s": 1.4345998351927847e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57b3508cd6fc6aa29315ab016cfa41efd39f857b57888e5ec91d29c3e2a3a23b
-size 834567
+oid sha256:cc51cd4f71940dccd6db05380852d2cdd97b4a97028c22941146cfd5fd3f4ea1
+size 836685