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": 77030.28172165212,
-      "required_pages_per_second_upper_estimate": 0.0053291336243823486
+      "required_token_positions_per_second": 76820.16249575745,
+      "required_pages_per_second_upper_estimate": 0.005314597088791181
     },
     {
       "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": 9628.785215206515,
-      "required_pages_per_second_upper_estimate": 0.004031371297707026
+      "required_token_positions_per_second": 9602.520311969682,
+      "required_pages_per_second_upper_estimate": 0.004020374731195313
     },
     {
       "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": 19257.57043041303,
-      "required_pages_per_second_upper_estimate": 0.004010844064370928
+      "required_token_positions_per_second": 19205.040623939363,
+      "required_pages_per_second_upper_estimate": 0.003999903490986618
     },
     {
       "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": 28886.355645619544,
-      "required_pages_per_second_upper_estimate": 0.0017316929140436938
+      "required_token_positions_per_second": 28807.560935909045,
+      "required_pages_per_second_upper_estimate": 0.0017269692915091047
     },
     {
       "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": 28886.355645619544,
-      "required_pages_per_second_upper_estimate": 0.002245787093796301
+      "required_token_positions_per_second": 28807.560935909045,
+      "required_pages_per_second_upper_estimate": 0.0022396611516976098
     },
     {
       "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": 28886.355645619544,
-      "required_pages_per_second_upper_estimate": 0.37052315446979317
+      "required_token_positions_per_second": 28807.560935909045,
+      "required_pages_per_second_upper_estimate": 0.3695124605367946
     }
   ],
   "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": 0.14308278833556187,
-    "producer_p95_seconds": 0.3753515740027069,
-    "gpu_update_seconds": 0.9571300838080163,
-    "producer_batches_per_second": 6.988960808163531,
-    "gpu_batches_per_second": 1.0447900624139013,
-    "producer_headroom": 6.689344644048503,
+    "producer_batch_seconds": 0.09717895390605236,
+    "producer_p95_seconds": 0.34301836299709976,
+    "gpu_update_seconds": 0.959748034952045,
+    "producer_batches_per_second": 10.290293935112214,
+    "gpu_batches_per_second": 1.0419401380175437,
+    "producer_headroom": 9.876089383302894,
     "sustainable": true,
-    "buffer_seconds": 60.29919527990503,
+    "buffer_seconds": 60.464126201978836,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301210.8686848358,
-    "upload_bytes_per_second": 5193992.536014112,
+    "generation_bytes_per_second": 321193.26128091797,
+    "upload_bytes_per_second": 5163385.273915293,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.63284058960728,
+    "estimated_full_shards_per_hour": 68.92059687443404,
     "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": 1.1129142319987295,
-      "bytes": 6226935,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.6370596409979044,
       "bytes": 13312554,
@@ -562,6 +556,12 @@
       "bytes": 5898529,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9206480690045282,
+      "bytes": 2670829,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -422,3 +422,4 @@
 {"step": 439300, "loss": 0.6133816242218018, "lr": 0.0003, "grad_norm": 0.1552689703390891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64300835505, "tokens_per_second": 152617.73701016037, "data_wait_s": 9.742005204316229e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 439400, "loss": 0.5345016121864319, "lr": 0.0003, "grad_norm": 0.14361488190581798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64315525897, "tokens_per_second": 151847.68204412117, "data_wait_s": 1.1209995136596262e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 439500, "loss": 0.5806330442428589, "lr": 0.0003, "grad_norm": 0.16768765683487677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64330335108, "tokens_per_second": 153885.36132112876, "data_wait_s": 8.455004717689008e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 439600, "loss": 0.6042528748512268, "lr": 0.0003, "grad_norm": 0.13606011704472593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64344771197, "tokens_per_second": 150769.85554022, "data_wait_s": 1.493600575486198e-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 sha256:324115719e03b1229bfeb2b436aca0df0a08965f7fae176ddef5f9dc7f86657d
-size 904461
+oid sha256:ee856eac6b7d0606bce223bf572d5ebfc464f14f39fb7655407919ec43aeabe2
+size 906579