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": 75933.83638621224,
-      "required_pages_per_second_upper_estimate": 0.0065817524854080425
+      "required_token_positions_per_second": 75420.67733792926,
+      "required_pages_per_second_upper_estimate": 0.006537273159692591
     },
     {
       "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": 9491.72954827653,
-      "required_pages_per_second_upper_estimate": 0.0037863928308108064
+      "required_token_positions_per_second": 9427.584667241157,
+      "required_pages_per_second_upper_estimate": 0.00376080447871436
     },
     {
       "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": 18983.45909655306,
-      "required_pages_per_second_upper_estimate": 0.004391489164518296
+      "required_token_positions_per_second": 18855.169334482314,
+      "required_pages_per_second_upper_estimate": 0.004361811585885928
     },
     {
       "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": 28475.18864482959,
-      "required_pages_per_second_upper_estimate": 0.0017339895907013706
+      "required_token_positions_per_second": 28282.75400172347,
+      "required_pages_per_second_upper_estimate": 0.0017222713305627523
     },
     {
       "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": 28475.18864482959,
-      "required_pages_per_second_upper_estimate": 0.0021920988462461348
+      "required_token_positions_per_second": 28282.75400172347,
+      "required_pages_per_second_upper_estimate": 0.0021772846947266397
     },
     {
       "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": 28475.18864482959,
-      "required_pages_per_second_upper_estimate": 0.35356654264287957
+      "required_token_positions_per_second": 28282.75400172347,
+      "required_pages_per_second_upper_estimate": 0.3511771484128223
     }
   ],
   "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": 0.11792124178913355,
-    "producer_p95_seconds": 0.384719423003844,
-    "gpu_update_seconds": 0.9709505473292172,
-    "producer_batches_per_second": 8.480236341033427,
-    "gpu_batches_per_second": 1.0299185707765333,
-    "producer_headroom": 8.233890116807524,
+    "producer_batch_seconds": 0.1482352566017653,
+    "producer_p95_seconds": 0.39227478799875826,
+    "gpu_update_seconds": 0.9775568531379656,
+    "producer_batches_per_second": 6.7460334533403525,
+    "gpu_batches_per_second": 1.0229584057336325,
+    "producer_headroom": 6.594631233810838,
     "sustainable": true,
-    "buffer_seconds": 60.198933934411464,
+    "buffer_seconds": 60.60852489455387,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373696.01402914454,
-    "upload_bytes_per_second": 3639816.6343356767,
+    "generation_bytes_per_second": 358977.1694439373,
+    "upload_bytes_per_second": 3698926.281376775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.18646541267158,
+    "estimated_full_shards_per_hour": 77.02814400185194,
     "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": 1.0518995679994987,
-      "bytes": 1242140,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.0469969259975187,
       "bytes": 7461003,
@@ -562,6 +556,12 @@
       "bytes": 6003121,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9798986519963364,
+      "bytes": 7781027,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 39300, "loss": 0.6332474946975708, "lr": 0.0003, "grad_norm": 0.13531782894139807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5750821886, "tokens_per_second": 149299.85389619958, "data_wait_s": 1.409299875376746e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 39400, "loss": 0.6049629338085651, "lr": 0.0003, "grad_norm": 0.12598470520687913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5765473879, "tokens_per_second": 134883.06262163044, "data_wait_s": 1.0157003998756409e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 39500, "loss": 0.6539227645844221, "lr": 0.0003, "grad_norm": 0.13698722597409166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5780151611, "tokens_per_second": 150391.8955422777, "data_wait_s": 8.754999726079404e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 39600, "loss": 0.6563158668577671, "lr": 0.0003, "grad_norm": 0.13424698124505394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5794733535, "tokens_per_second": 148580.15477793117, "data_wait_s": 1.2144999345764518e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cee839750e44f7ab04f4c14349d2fd7a9d0c1020e7566e730f1bf74e3bca951b
-size 836482
+oid sha256:4f80595d524937eea2048d4b0cd63b46ea387229869f80b5d31f9b9cfc634948
+size 838600