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": 76214.30466202498,
-      "required_pages_per_second_upper_estimate": 0.005272682438069715
+      "required_token_positions_per_second": 75457.90810745048,
+      "required_pages_per_second_upper_estimate": 0.005220353169342442
     },
     {
       "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": 9526.788082753123,
-      "required_pages_per_second_upper_estimate": 0.003988667228291121
+      "required_token_positions_per_second": 9432.23851343131,
+      "required_pages_per_second_upper_estimate": 0.003949081298035603
     },
     {
       "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": 19053.576165506245,
-      "required_pages_per_second_upper_estimate": 0.003968357438681379
+      "required_token_positions_per_second": 18864.47702686262,
+      "required_pages_per_second_upper_estimate": 0.003928973074981551
     },
     {
       "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": 28580.364248259368,
-      "required_pages_per_second_upper_estimate": 0.001713349196993762
+      "required_token_positions_per_second": 28296.71554029393,
+      "required_pages_per_second_upper_estimate": 0.0016963448900577335
     },
     {
       "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": 28580.364248259368,
-      "required_pages_per_second_upper_estimate": 0.0022219976085654635
+      "required_token_positions_per_second": 28296.71554029393,
+      "required_pages_per_second_upper_estimate": 0.002199945169159963
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28580.364248259368,
-      "required_pages_per_second_upper_estimate": 0.3665982253724217
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28296.71554029393,
+      "required_pages_per_second_upper_estimate": 0.3629598843048952
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10846716339102613,
-    "producer_p95_seconds": 0.29848488599964185,
-    "gpu_update_seconds": 0.9673774539694275,
-    "producer_batches_per_second": 9.219380029281133,
-    "gpu_batches_per_second": 1.0337226652292886,
-    "producer_headroom": 8.918620379902569,
+    "producer_batch_seconds": 0.1096222646174283,
+    "producer_p95_seconds": 0.29750585200235946,
+    "gpu_update_seconds": 0.9770745286897282,
+    "producer_batches_per_second": 9.122234461128027,
+    "gpu_batches_per_second": 1.0234633803636404,
+    "producer_headroom": 8.913102936703863,
     "sustainable": true,
-    "buffer_seconds": 60.944779600073936,
+    "buffer_seconds": 60.57862077876315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422676.3614830114,
-    "upload_bytes_per_second": 4898378.845196422,
+    "generation_bytes_per_second": 418344.62579474493,
+    "upload_bytes_per_second": 4850078.272175883,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.69650777213818,
+    "estimated_full_shards_per_hour": 89.76701813108217,
     "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": 1.1877205400014645,
-      "bytes": 6194082,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0285773200012045,
       "bytes": 923291,
@@ -562,6 +556,12 @@
       "bytes": 11872388,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.3592714799960959,
+      "bytes": 2722837,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 793400, "loss": 0.5702718496322632, "lr": 3e-05, "grad_norm": 0.2712648949898607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116131171122, "tokens_per_second": 151034.46029946272, "data_wait_s": 9.796996891964227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 793500, "loss": 0.5302921533584595, "lr": 3e-05, "grad_norm": 0.24440704053070145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116145981237, "tokens_per_second": 153521.0138528239, "data_wait_s": 1.4961005945224315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 793600, "loss": 0.5755289793014526, "lr": 3e-05, "grad_norm": 0.3006565562961368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116160663144, "tokens_per_second": 152079.74165767743, "data_wait_s": 1.105800038203597e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 793700, "loss": 0.5079626441001892, "lr": 3e-05, "grad_norm": 0.2647570273944449, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116175177861, "tokens_per_second": 151070.3173296538, "data_wait_s": 1.0025003575719893e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbe06dcf95b0f34ab5138aeeafe7e6ae129cac8e96c51e6aa25d2afaca246ffb
-size 716511
+oid sha256:597371d5fddeb6f4e1ab9e9eee8dbe73a87203a7ce4b8b6e9bdf2096e9213b95
+size 718629