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": 76619.1211392568,
-      "required_pages_per_second_upper_estimate": 0.00530068858126825
+      "required_token_positions_per_second": 76666.71129443996,
+      "required_pages_per_second_upper_estimate": 0.005303980978628195
     },
     {
       "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": 9577.3901424071,
-      "required_pages_per_second_upper_estimate": 0.0040098532539770745
+      "required_token_positions_per_second": 9583.338911804994,
+      "required_pages_per_second_upper_estimate": 0.0040123438794995425
     },
     {
       "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": 19154.7802848142,
-      "required_pages_per_second_upper_estimate": 0.0039894355878011215
+      "required_token_positions_per_second": 19166.67782360999,
+      "required_pages_per_second_upper_estimate": 0.003991913531373088
     },
     {
       "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": 28732.1704272213,
-      "required_pages_per_second_upper_estimate": 0.0017224497456279158
+      "required_token_positions_per_second": 28750.016735414985,
+      "required_pages_per_second_upper_estimate": 0.0017235196045543926
     },
     {
       "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": 28732.1704272213,
-      "required_pages_per_second_upper_estimate": 0.00223379987125494
+      "required_token_positions_per_second": 28750.016735414985,
+      "required_pages_per_second_upper_estimate": 0.0022351873432193852
     },
     {
       "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": 28732.1704272213,
-      "required_pages_per_second_upper_estimate": 0.3685454320393697
+      "required_token_positions_per_second": 28750.016735414985,
+      "required_pages_per_second_upper_estimate": 0.3687743453190055
     }
   ],
   "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": 0.10972266731235436,
-    "producer_p95_seconds": 0.2201323980007146,
-    "gpu_update_seconds": 0.962266323389404,
-    "producer_batches_per_second": 9.11388707998902,
-    "gpu_batches_per_second": 1.039213340104937,
-    "producer_headroom": 8.769986612247225,
+    "producer_batch_seconds": 0.23922059314068633,
+    "producer_p95_seconds": 0.23176129599960404,
+    "gpu_update_seconds": 0.9616690054285257,
+    "producer_batches_per_second": 4.180242122432566,
+    "gpu_batches_per_second": 1.0398588228955072,
+    "producer_headroom": 4.020009284330155,
     "sustainable": true,
-    "buffer_seconds": 60.62277837353245,
+    "buffer_seconds": 60.58514734199712,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439002.49802355544,
-    "upload_bytes_per_second": 5283536.887516052,
+    "generation_bytes_per_second": 442640.3784859692,
+    "upload_bytes_per_second": 5360234.299301244,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.19971662073132,
+    "estimated_full_shards_per_hour": 94.98032108244236,
     "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": 1.3408090630000515,
-      "bytes": 7658166,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8867853440006002,
       "bytes": 5785267,
@@ -562,6 +556,12 @@
       "bytes": 6125180,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2460403479999513,
+      "bytes": 13328563,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -176,3 +176,4 @@
 {"step": 622000, "loss": 0.5392137169837952, "lr": 0.0001, "grad_norm": 0.18700737955620458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91044280356, "tokens_per_second": 152117.30410978638, "data_wait_s": 9.61499972618185e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 622100, "loss": 0.6104187369346619, "lr": 0.0001, "grad_norm": 0.1739834834462838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91059036809, "tokens_per_second": 153491.74303231735, "data_wait_s": 1.4015997294336557e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 622200, "loss": 0.55616295337677, "lr": 0.0001, "grad_norm": 0.16853990272410171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91073640461, "tokens_per_second": 152260.05171838557, "data_wait_s": 1.0214997018920258e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 622300, "loss": 0.515613317489624, "lr": 0.0001, "grad_norm": 0.15178274193613836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91088102509, "tokens_per_second": 149590.18571085605, "data_wait_s": 1.367799995932728e-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 sha256:36ec98ba1439242c0791c6600958a40947787e1146c640df5cc1ae380f459984
-size 380301
+oid sha256:11fce2a46da5670e708b514d735f6d84dd8f8016748217d96a0c1965ec56a2e4
+size 382419