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": 76135.25236083151,
-      "required_pages_per_second_upper_estimate": 0.005267213416446541
+      "required_token_positions_per_second": 75577.74094653316,
+      "required_pages_per_second_upper_estimate": 0.0052286434832006755
     },
     {
       "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": 9516.906545103939,
-      "required_pages_per_second_upper_estimate": 0.003984530034827378
+      "required_token_positions_per_second": 9447.217618316645,
+      "required_pages_per_second_upper_estimate": 0.003955352736451814
     },
     {
       "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": 19033.813090207877,
-      "required_pages_per_second_upper_estimate": 0.003964241311284073
+      "required_token_positions_per_second": 18894.43523663329,
+      "required_pages_per_second_upper_estimate": 0.003935212580025662
     },
     {
       "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": 28550.719635311812,
-      "required_pages_per_second_upper_estimate": 0.001711572047712259
+      "required_token_positions_per_second": 28341.652854949934,
+      "required_pages_per_second_upper_estimate": 0.0016990388134560544
     },
     {
       "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": 28550.719635311812,
-      "required_pages_per_second_upper_estimate": 0.0022196928703016633
+      "required_token_positions_per_second": 28341.652854949934,
+      "required_pages_per_second_upper_estimate": 0.002203438847716109
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28550.719635311812,
-      "required_pages_per_second_upper_estimate": 0.36621797610743595
+      "required_token_positions_per_second": 28341.652854949934,
+      "required_pages_per_second_upper_estimate": 0.3635362919273731
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09514616663281572,
-    "producer_p95_seconds": 0.27275808500053245,
-    "gpu_update_seconds": 0.968381895558411,
-    "producer_batches_per_second": 10.510144921120785,
-    "gpu_batches_per_second": 1.0326504497725628,
-    "producer_headroom": 10.177834061308552,
+    "producer_batch_seconds": 0.13471599394551959,
+    "producer_p95_seconds": 0.3085167899989756,
+    "gpu_update_seconds": 0.975525321035439,
+    "producer_batches_per_second": 7.423023582518415,
+    "gpu_batches_per_second": 1.025088717265261,
+    "producer_headroom": 7.2413474633899115,
     "sustainable": true,
-    "buffer_seconds": 60.03967752462148,
+    "buffer_seconds": 60.48256990419722,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300784.76653921173,
-    "upload_bytes_per_second": 5422978.02601734,
+    "generation_bytes_per_second": 321093.0644514783,
+    "upload_bytes_per_second": 5434370.47019392,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.54140898830666,
+    "estimated_full_shards_per_hour": 68.8990969672991,
     "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": 1.5744073349997052,
-      "bytes": 1306429,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.5331175120008993,
       "bytes": 11738494,
@@ -562,6 +556,12 @@
       "bytes": 6142168,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3281040260008012,
+      "bytes": 876378,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 746200, "loss": 0.5053272843360901, "lr": 5e-05, "grad_norm": 0.23429421098434564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109223438106, "tokens_per_second": 149215.56514957664, "data_wait_s": 9.994000720325857e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 746300, "loss": 0.540273129940033, "lr": 5e-05, "grad_norm": 0.24720239149437304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109238001258, "tokens_per_second": 150270.01715989434, "data_wait_s": 1.0635998478392139e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 746400, "loss": 0.5150867700576782, "lr": 5e-05, "grad_norm": 0.2449501627606696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109252631258, "tokens_per_second": 150056.3392223231, "data_wait_s": 1.0732001101132482e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 746500, "loss": 0.535110592842102, "lr": 5e-05, "grad_norm": 0.2323000257424983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109267363758, "tokens_per_second": 151393.53563973712, "data_wait_s": 1.2826003512600437e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b0ee230b20a0ab124a8402b4ccf5359ba6fc025c125ad36724dd8304cd58d3b5
-size 563232
+oid sha256:905cb8c1048f98cc8e040fbb85b360cc6011a9fe8dc63a1d980161e9a6aaf2f2
+size 565350