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": 76220.68501750415,
-      "required_pages_per_second_upper_estimate": 0.006606615797757216
+      "required_token_positions_per_second": 76127.77658254562,
+      "required_pages_per_second_upper_estimate": 0.006598562730089286
     },
     {
       "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": 9527.585627188018,
-      "required_pages_per_second_upper_estimate": 0.003800696356784753
+      "required_token_positions_per_second": 9515.972072818202,
+      "required_pages_per_second_upper_estimate": 0.0037960635363085216
     },
     {
       "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": 19055.171254376037,
-      "required_pages_per_second_upper_estimate": 0.00440807851013977
+      "required_token_positions_per_second": 19031.944145636404,
+      "required_pages_per_second_upper_estimate": 0.004402705327316008
     },
     {
       "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": 28582.756881564055,
-      "required_pages_per_second_upper_estimate": 0.001740539931951578
+      "required_token_positions_per_second": 28547.916218454604,
+      "required_pages_per_second_upper_estimate": 0.0017384183183630445
     },
     {
       "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": 28582.756881564055,
-      "required_pages_per_second_upper_estimate": 0.0022003797468778913
+      "required_token_positions_per_second": 28547.916218454604,
+      "required_pages_per_second_upper_estimate": 0.002197697616186597
     },
     {
       "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": 28582.756881564055,
-      "required_pages_per_second_upper_estimate": 0.3549021801353919
+      "required_token_positions_per_second": 28547.916218454604,
+      "required_pages_per_second_upper_estimate": 0.35446957570377097
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3607578066562951,
-    "producer_p95_seconds": 0.26255481299995154,
-    "gpu_update_seconds": 0.9672964757935237,
-    "producer_batches_per_second": 2.771942786958815,
-    "gpu_batches_per_second": 1.0338092043389777,
-    "producer_headroom": 2.6812904889265403,
+    "producer_batch_seconds": 0.09582121260153542,
+    "producer_p95_seconds": 0.27117204399837647,
+    "gpu_update_seconds": 0.9684769910501257,
+    "producer_batches_per_second": 10.436102537737831,
+    "gpu_batches_per_second": 1.0325490530401695,
+    "producer_headroom": 10.107125184038916,
     "sustainable": true,
-    "buffer_seconds": 60.939677974991994,
+    "buffer_seconds": 60.0455734451078,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417326.9139662742,
-    "upload_bytes_per_second": 6222851.788199006,
+    "generation_bytes_per_second": 416734.52252620424,
+    "upload_bytes_per_second": 6385147.086400999,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.5486408638112,
+    "estimated_full_shards_per_hour": 89.42152745093914,
     "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": 0.9305132150002464,
-      "bytes": 1095891,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.8200730230000772,
       "bytes": 7217811,
@@ -562,6 +556,12 @@
       "bytes": 1074472,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.0881272910010011,
+      "bytes": 13091178,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 54700, "loss": 0.6260981969535351, "lr": 0.0003, "grad_norm": 0.13590119837122686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8005840156, "tokens_per_second": 148382.28499399193, "data_wait_s": 7.453998478013091e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 54800, "loss": 0.6338471099734306, "lr": 0.0003, "grad_norm": 0.13873746194503214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8020481174, "tokens_per_second": 150470.1087460764, "data_wait_s": 1.324500044574961e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 54900, "loss": 0.6329132784157991, "lr": 0.0003, "grad_norm": 0.1287076918518521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8035088021, "tokens_per_second": 149553.46658519574, "data_wait_s": 1.0747000487754121e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 55000, "loss": 0.536142909899354, "lr": 0.0003, "grad_norm": 0.15281186596519372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8049776223, "tokens_per_second": 150058.34292751199, "data_wait_s": 9.393001164426096e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4882a5542fd16042e9d5769ae64efa7a22504470694fec0f3c8c0cfd9450e5a7
-size 317313
+oid sha256:049faeef7c761d86eea6e719327e3f3a04c08c224a9f9586cd2c415ef09fc47a
+size 319431