Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75820.56224056795,
-      "required_pages_per_second_upper_estimate": 0.006571934169554293
+      "required_token_positions_per_second": 75243.30017740611,
+      "required_pages_per_second_upper_estimate": 0.0065218985569767485
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9477.570280070993,
-      "required_pages_per_second_upper_estimate": 0.003780744487023693
+      "required_token_positions_per_second": 9405.412522175764,
+      "required_pages_per_second_upper_estimate": 0.003751959678544664
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18955.140560141986,
-      "required_pages_per_second_upper_estimate": 0.004384938169508814
+      "required_token_positions_per_second": 18810.825044351528,
+      "required_pages_per_second_upper_estimate": 0.004351553314797022
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28432.71084021298,
-      "required_pages_per_second_upper_estimate": 0.0017314029152640362
+      "required_token_positions_per_second": 28216.237566527292,
+      "required_pages_per_second_upper_estimate": 0.0017182208286440695
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28432.71084021298,
-      "required_pages_per_second_upper_estimate": 0.0021888287872606593
+      "required_token_positions_per_second": 28216.237566527292,
+      "required_pages_per_second_upper_estimate": 0.0021721640754159602
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28432.71084021298,
-      "required_pages_per_second_upper_estimate": 0.3530391104736082
+      "required_token_positions_per_second": 28216.237566527292,
+      "required_pages_per_second_upper_estimate": 0.3503512369038739
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11350576055460238,
-    "producer_p95_seconds": 0.33627448399784043,
-    "gpu_update_seconds": 0.9724011247248663,
-    "producer_batches_per_second": 8.81012554000681,
-    "gpu_batches_per_second": 1.028382191847981,
-    "producer_headroom": 8.566975984069893,
+    "producer_batch_seconds": 0.36608840310950086,
+    "producer_p95_seconds": 0.35924587000044994,
+    "gpu_update_seconds": 0.9798613275356957,
+    "producer_batches_per_second": 2.7315806551263235,
+    "gpu_batches_per_second": 1.0205525740208077,
+    "producer_headroom": 2.6765702470029047,
     "sustainable": true,
-    "buffer_seconds": 60.28886973294171,
+    "buffer_seconds": 60.751402307213134,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303205.0528035812,
-    "upload_bytes_per_second": 4459818.2917380165,
+    "generation_bytes_per_second": 317686.3158077866,
+    "upload_bytes_per_second": 4559177.617897854,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.06074607925964,
+    "estimated_full_shards_per_hour": 68.16808801341246,
     "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": 0.8550500759993156,
-      "bytes": 2622896,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1337759699999879,
       "bytes": 5871486,
@@ -562,6 +556,12 @@
       "bytes": 913952,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.163448106002761,
+      "bytes": 13378237,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 184500, "loss": 0.5798609592020512, "lr": 0.0003, "grad_norm": 0.1502639114966204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27003277715, "tokens_per_second": 147708.54063929812, "data_wait_s": 1.5797999367350712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184600, "loss": 0.5771688465029001, "lr": 0.0003, "grad_norm": 0.13869490845219812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27017745657, "tokens_per_second": 148701.86428663874, "data_wait_s": 8.061000698944554e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 184700, "loss": 0.5921379346400499, "lr": 0.0003, "grad_norm": 0.1538417757211502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27032382142, "tokens_per_second": 148764.65570608995, "data_wait_s": 1.221599814016372e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 184800, "loss": 0.6112805008888245, "lr": 0.0003, "grad_norm": 0.12806757929661203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27046955426, "tokens_per_second": 149143.50961878238, "data_wait_s": 8.838000212563202e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b4a3e34ace1987d06f951d9f732e7603c28e7666dc25d8f8832cc304b2000c2
-size 633126
+oid sha256:89852239424076a62c1ea1c383b6d4365bd86378dbffe755ead2b084fdb44bbf
+size 635244