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": 75527.55228796769,
-      "required_pages_per_second_upper_estimate": 0.0065465368094898375
+      "required_token_positions_per_second": 75482.0224017691,
+      "required_pages_per_second_upper_estimate": 0.006542590394348587
     },
     {
       "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": 9440.94403599596,
-      "required_pages_per_second_upper_estimate": 0.0037661337306510137
+      "required_token_positions_per_second": 9435.252800221138,
+      "required_pages_per_second_upper_estimate": 0.003763863411608879
     },
     {
       "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": 18881.88807199192,
-      "required_pages_per_second_upper_estimate": 0.004367992495575056
+      "required_token_positions_per_second": 18870.505600442277,
+      "required_pages_per_second_upper_estimate": 0.004365359361106703
     },
     {
       "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": 28322.832107987884,
-      "required_pages_per_second_upper_estimate": 0.0017247118769606823
+      "required_token_positions_per_second": 28305.758400663417,
+      "required_pages_per_second_upper_estimate": 0.0017236721777634414
     },
     {
       "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": 28322.832107987884,
-      "required_pages_per_second_upper_estimate": 0.002180370018290173
+      "required_token_positions_per_second": 28305.758400663417,
+      "required_pages_per_second_upper_estimate": 0.002179055637037289
     },
     {
       "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": 28322.832107987884,
-      "required_pages_per_second_upper_estimate": 0.3516747843598332
+      "required_token_positions_per_second": 28305.758400663417,
+      "required_pages_per_second_upper_estimate": 0.35146278605688586
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05511853775780651,
-    "producer_p95_seconds": 0.14779246800026158,
-    "gpu_update_seconds": 0.9761735653618636,
-    "producer_batches_per_second": 18.14271641954741,
-    "gpu_batches_per_second": 1.024407990016923,
-    "producer_headroom": 17.71044017261882,
+    "producer_batch_seconds": 0.06234097357031487,
+    "producer_p95_seconds": 0.14544689100011965,
+    "gpu_update_seconds": 0.9767623820088848,
+    "producer_batches_per_second": 16.04081461564106,
+    "gpu_batches_per_second": 1.023790451412884,
+    "producer_headroom": 15.668064293336498,
     "sustainable": true,
-    "buffer_seconds": 60.52276105243554,
+    "buffer_seconds": 60.55926768455085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -22,3 +22,4 @@
 {"step": 197200, "loss": 0.5887098088860512, "lr": 0.0003, "grad_norm": 0.13839620821846402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28862254588, "tokens_per_second": 150090.3536678603, "data_wait_s": 1.8150999949284596e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197300, "loss": 0.6105303391814232, "lr": 0.0003, "grad_norm": 0.14922885645554973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28876925716, "tokens_per_second": 150022.3374675087, "data_wait_s": 1.7724000372254523e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197400, "loss": 0.6115903630852699, "lr": 0.0003, "grad_norm": 0.14707520080594957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28891607788, "tokens_per_second": 149733.96647133294, "data_wait_s": 1.8290000298293307e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 197500, "loss": 0.5895135682076216, "lr": 0.0003, "grad_norm": 0.13332616485866206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28906201905, "tokens_per_second": 148764.0805458956, "data_wait_s": 1.8090000139636686e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0825837c7dcb9ab09c4c9b4bef04b66cefdc0e1c71d04e7f6aaa9ef447b6870
-size 50997
+oid sha256:79c34919a04f66e4575a2c559e9a43c836233257b7a1c73614dc09329467b0fd
+size 53115