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": 76336.14723027711,
-      "required_pages_per_second_upper_estimate": 0.006616623769724074
+      "required_token_positions_per_second": 74683.81509213088,
+      "required_pages_per_second_upper_estimate": 0.006473403807787076
     },
     {
       "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": 9542.01840378464,
-      "required_pages_per_second_upper_estimate": 0.003806453807158385
+      "required_token_positions_per_second": 9335.47688651636,
+      "required_pages_per_second_upper_estimate": 0.0037240613078491944
     },
     {
       "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": 19084.03680756928,
-      "required_pages_per_second_upper_estimate": 0.004414756047854642
+      "required_token_positions_per_second": 18670.95377303272,
+      "required_pages_per_second_upper_estimate": 0.0043191965578277145
     },
     {
       "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": 28626.05521135392,
-      "required_pages_per_second_upper_estimate": 0.0017431765730670007
+      "required_token_positions_per_second": 28006.43065954908,
+      "required_pages_per_second_upper_estimate": 0.0017054446887808652
     },
     {
       "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": 28626.05521135392,
-      "required_pages_per_second_upper_estimate": 0.002203712972162563
+      "required_token_positions_per_second": 28006.43065954908,
+      "required_pages_per_second_upper_estimate": 0.0021560125589340875
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28626.05521135392,
-      "required_pages_per_second_upper_estimate": 0.355439800481194
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28006.43065954908,
+      "required_pages_per_second_upper_estimate": 0.34774613729775233
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11260409534372684,
-    "producer_p95_seconds": 0.36939199000335066,
-    "gpu_update_seconds": 0.9658333918476482,
-    "producer_batches_per_second": 8.88067167492865,
-    "gpu_batches_per_second": 1.0353752608273263,
-    "producer_headroom": 8.577249245681672,
+    "producer_batch_seconds": 0.39822720660953337,
+    "producer_p95_seconds": 0.35639980899577495,
+    "gpu_update_seconds": 0.9872018443226048,
+    "producer_batches_per_second": 2.5111292834909498,
+    "gpu_batches_per_second": 1.012964071887626,
+    "producer_headroom": 2.478991459994767,
     "sustainable": true,
-    "buffer_seconds": 60.847503686401836,
+    "buffer_seconds": 60.21931250367889,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488900.51708337123,
-    "upload_bytes_per_second": 5129809.9256916335,
+    "generation_bytes_per_second": 421554.67457388015,
+    "upload_bytes_per_second": 5132631.865739546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.9066699445329,
+    "estimated_full_shards_per_hour": 90.45581987297348,
     "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": 1.1386810379990493,
-      "bytes": 6058898,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.2615420420006558,
       "bytes": 12864855,
@@ -562,6 +556,12 @@
       "bytes": 8604864,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.0984054020009353,
+      "bytes": 6085035,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -365,3 +365,4 @@
 {"step": 191500, "loss": 0.5999083891510963, "lr": 0.0003, "grad_norm": 0.15876979244002104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28028087206, "tokens_per_second": 149408.16969388092, "data_wait_s": 9.577001037541777e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 191600, "loss": 0.6017245966941118, "lr": 0.0003, "grad_norm": 0.1600590673131796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28042533223, "tokens_per_second": 148528.88133725265, "data_wait_s": 1.0511001164559275e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 191700, "loss": 0.5890930593013763, "lr": 0.0003, "grad_norm": 0.1589695955741389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28057264346, "tokens_per_second": 151062.45055904746, "data_wait_s": 1.0711002687457949e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 191800, "loss": 0.5835154447704554, "lr": 0.0003, "grad_norm": 0.1517761281663549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28071917953, "tokens_per_second": 150295.6748934891, "data_wait_s": 8.285001968033612e-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:b65b3b8e830479cacfac0000198574a037bf45fc5824e4978c806961e128739b
-size 782952
+oid sha256:9a9394d84d73fe8994fa35aeef82ea23787dbf24042af46801a382ca3284af68
+size 785070