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": 76467.05579289528,
-      "required_pages_per_second_upper_estimate": 0.005290168347766745
+      "required_token_positions_per_second": 76330.7444726351,
+      "required_pages_per_second_upper_estimate": 0.005280738014345306
     },
     {
       "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": 9558.38197411191,
-      "required_pages_per_second_upper_estimate": 0.00400189493084757
+      "required_token_positions_per_second": 9541.343059079387,
+      "required_pages_per_second_upper_estimate": 0.003994761092936459
     },
     {
       "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": 19116.76394822382,
-      "required_pages_per_second_upper_estimate": 0.003981517787447561
+      "required_token_positions_per_second": 19082.686118158774,
+      "required_pages_per_second_upper_estimate": 0.003974420274137825
     },
     {
       "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": 28675.145922335727,
-      "required_pages_per_second_upper_estimate": 0.0017190312136314031
+      "required_token_positions_per_second": 28624.02917723816,
+      "required_pages_per_second_upper_estimate": 0.0017159668428135546
     },
     {
       "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": 28675.145922335727,
-      "required_pages_per_second_upper_estimate": 0.0022293664668243756
+      "required_token_positions_per_second": 28624.02917723816,
+      "required_pages_per_second_upper_estimate": 0.00222539236473184
     },
     {
       "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": 28675.145922335727,
-      "required_pages_per_second_upper_estimate": 0.36781398291884054
+      "required_token_positions_per_second": 28624.02917723816,
+      "required_pages_per_second_upper_estimate": 0.3671583121976137
     }
   ],
   "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": 0.524201496164153,
-    "producer_p95_seconds": 0.366478198004188,
-    "gpu_update_seconds": 0.9641799234390065,
-    "producer_batches_per_second": 1.9076633838657555,
-    "gpu_batches_per_second": 1.0371508218437402,
-    "producer_headroom": 1.8393307354030801,
+    "producer_batch_seconds": 0.11743164354680857,
+    "producer_p95_seconds": 0.3687620260025142,
+    "gpu_update_seconds": 0.9659017543898293,
+    "producer_batches_per_second": 8.51559230371665,
+    "gpu_batches_per_second": 1.0353019812369126,
+    "producer_headroom": 8.22522554582844,
     "sustainable": true,
-    "buffer_seconds": 60.74333517665741,
+    "buffer_seconds": 60.85181052655925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296066.99378570984,
-    "upload_bytes_per_second": 5252441.806014503,
+    "generation_bytes_per_second": 295262.3209396798,
+    "upload_bytes_per_second": 5277823.378368886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.52908477953407,
+    "estimated_full_shards_per_hour": 63.356420718601186,
     "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": 0.9928698779986007,
-      "bytes": 5905666,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 2.588622392999241,
       "bytes": 7185173,
@@ -562,6 +556,12 @@
       "bytes": 5955041,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.6710909179964801,
+      "bytes": 11523305,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -412,3 +412,4 @@
 {"step": 481400, "loss": 0.576505720615387, "lr": 0.0003, "grad_norm": 0.16806699233961728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70461030959, "tokens_per_second": 148648.06692966688, "data_wait_s": 9.504998160991818e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 481500, "loss": 0.5666149854660034, "lr": 0.0003, "grad_norm": 0.1480821603432525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70475622780, "tokens_per_second": 152071.17167110345, "data_wait_s": 1.0850002581719309e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 481600, "loss": 0.6309847831726074, "lr": 0.0003, "grad_norm": 0.18463335991846513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70490304058, "tokens_per_second": 152411.35688619746, "data_wait_s": 1.5852994692977518e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 481700, "loss": 0.5596342086791992, "lr": 0.0003, "grad_norm": 0.1570385359003396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70504874513, "tokens_per_second": 151225.64996765272, "data_wait_s": 1.6231999325100332e-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 sha256:1402433dfc72fd17dfc7ee88ab7b18aa96814091350fd291ab91db13fb2eba7b
-size 883281
+oid sha256:773ccc9d916c13b3c06f2fdb69b0d809c7e183c1d0013ce97c9963f0d0b657f2
+size 885399