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": 76810.26986814145,
-      "required_pages_per_second_upper_estimate": 0.005313912693858647
+      "required_token_positions_per_second": 76812.1644317684,
+      "required_pages_per_second_upper_estimate": 0.005314043764166353
     },
     {
       "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": 9601.283733517681,
-      "required_pages_per_second_upper_estimate": 0.0040198570016201545
+      "required_token_positions_per_second": 9601.52055397105,
+      "required_pages_per_second_upper_estimate": 0.004019956153398607
     },
     {
       "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": 19202.567467035362,
-      "required_pages_per_second_upper_estimate": 0.003999388397625048
+      "required_token_positions_per_second": 19203.0411079421,
+      "required_pages_per_second_upper_estimate": 0.00399948704453517
     },
     {
       "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": 28803.851200553047,
-      "required_pages_per_second_upper_estimate": 0.0017267468985389506
+      "required_token_positions_per_second": 28804.561661913147,
+      "required_pages_per_second_upper_estimate": 0.0017267894896126768
     },
     {
       "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": 28803.851200553047,
-      "required_pages_per_second_upper_estimate": 0.002239372736091082
+      "required_token_positions_per_second": 28804.561661913147,
+      "required_pages_per_second_upper_estimate": 0.002239427971336841
     },
     {
       "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": 28803.851200553047,
-      "required_pages_per_second_upper_estimate": 0.36946487603485134
+      "required_token_positions_per_second": 28804.561661913147,
+      "required_pages_per_second_upper_estimate": 0.36947398907034473
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1594818081641165,
-    "producer_p95_seconds": 0.38034316500124987,
-    "gpu_update_seconds": 0.9598716438123089,
-    "producer_batches_per_second": 6.270307638918535,
-    "gpu_batches_per_second": 1.0418059606681513,
-    "producer_headroom": 6.018690500577612,
+    "producer_batch_seconds": 0.10191386870337737,
+    "producer_p95_seconds": 0.3661101939942455,
+    "gpu_update_seconds": 0.9598479686832933,
+    "producer_batches_per_second": 9.812207236588405,
+    "gpu_batches_per_second": 1.0418316573319282,
+    "producer_headroom": 9.41822718433889,
     "sustainable": true,
-    "buffer_seconds": 60.471913560175466,
+    "buffer_seconds": 60.470422027047476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366804.33146936534,
+    "generation_bytes_per_second": 366783.0225817086,
     "upload_bytes_per_second": 4435833.203760793,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.70767076550217,
+    "estimated_full_shards_per_hour": 78.70309837425656,
     "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": [

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

@@ -433,3 +433,4 @@
 {"step": 647700, "loss": 0.5617843270301819, "lr": 0.0001, "grad_norm": 0.18590474809886204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94806559444, "tokens_per_second": 150624.9594513292, "data_wait_s": 1.0046001989394426e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647800, "loss": 0.5408211946487427, "lr": 0.0001, "grad_norm": 0.17289111194633108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94821128922, "tokens_per_second": 149982.78966636682, "data_wait_s": 1.764699845807627e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647900, "loss": 0.5759742856025696, "lr": 0.0001, "grad_norm": 0.19641977747136147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94835801434, "tokens_per_second": 152325.6590805752, "data_wait_s": 1.2803000572603196e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 648000, "loss": 0.5907433032989502, "lr": 0.0001, "grad_norm": 0.16774510031635662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94850495306, "tokens_per_second": 152423.47781005746, "data_wait_s": 1.1436000932008028e-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 sha256:23163b02345bd0b840058d2473153e4267646571fbb2595a4173efee6781f9cd
-size 928542
+oid sha256:68ad5169c1b40bb0391b13387c008ca5df007fb3efb815bcaca89ebc0ac024bb
+size 930660